ryubing-mcp
This server controls the Ryubing Nintendo Switch emulator via MCP, enabling session management, input simulation, monitoring, and diagnostics. Capabilities include:
Game launching: Launch user-owned game files (.xci, .nsp, .nca, .nro, or unpacked directories) with optional updates.
Session management: Create isolated sessions with separate profiles, list active sessions, stop sessions, and clean up unused profiles.
Controller input: Press digital buttons (including D-pad, stick clicks, etc.), set analog stick coordinates, and send touchscreen gestures (tap, long press, swipe) in handheld or docked modes for players 1–8.
Screenshots: Capture the emulator's rendered frame as PNG or JPEG.
Logs: Retrieve structured emulator logs with cursor-based pagination and level filtering (trace through critical).
Status monitoring: Query emulator state, title info, frame dimensions, and bridge compatibility.
Environment validation: Check executable, profile, keys, firmware, and game paths.
Scripted input: Execute sequences of input operations with validation.
Frame sequences: Record bounded series of frames for visual-loop evidence.
Diagnostics: Generate and export diagnostic reports, inspect patches, clean up artifacts, and get runtime metrics and advice.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ryubing-mcptake a screenshot of the current game"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ryubing-mcp
ryubing-mcp is a local Model Context Protocol (MCP) server for controlling the
Ryubing Nintendo Switch emulator.
It lets an MCP client start isolated emulator sessions, launch a user-owned
game, send controller input, capture the rendered frame, and inspect logs.
Use theRyubing MCP Bridge custom build to use the complete MCP feature set. An unmodified Ryubing executable works through the Windows patchless compatibility backend, but analog-stick and touchscreen tools are intentionally unavailable there.
Only use game, update, firmware, and key files that you are legally entitled to use. This project does not distribute any of them.
What is included
A stdio MCP server for local MCP clients.
Managed Ryubing processes with one retained copied profile per title by default.
A full control-bridge backend for the recommended custom Ryubing build.
A Windows patchless fallback for an unmodified Ryubing executable.
Optional game-path confinement.
Cursor-based log reading and renderer screenshots.
Persistent per-session manifests and NDJSON logs, including after a managed process exits.
Installation/environment checks that do not expose key contents.
Read-only launch and patch preflight checks.
A versioned capability contract exposed as both an MCP tool and resource so clients can discover bridge/patchless limitations without guessing.
One-call diagnostic reports with repeated-log and gameplay-failure hints, exportable bundles, and bounded frame sequences for visual-loop evidence.
Read-only font-atlas review harness for canvas/alpha/pixel-format, glyph rectangles, clipping, metric drift, and translated-text glyph coverage.
A repository-bundled
nsw-workspace-cleanupskill that correlates Ryubing and Eden handoffs, manifests, sessions, and artifacts into a read-only cleanup plan.Automated tests and live smoke-test scripts.
Related MCP server: algorithmaide-mcp
Choose a backend
Capability | Custom build ( | Unmodified build ( |
Windows managed sessions | Yes | Yes |
Status and current title | Yes | Yes, inferred from the process/window |
Digital buttons | Players 1-8 and handheld | Player 1 or handheld keyboard bindings |
Analog sticks | Yes | Pending / unavailable |
Touch tap, hold, and swipe | Yes | Pending / unavailable |
Screenshots | PNG and JPEG renderer frames | Native PNG through the configured hotkey |
Logs | Structured in-memory bridge logs | Parsed per-session file logs |
Base game and session update launch | Yes | Yes |
Window focus/hotkey dependency | No | Yes |
Recommended for complete MCP use | Yes | Compatibility fallback only |
backend: "auto" selects patchless managed sessions on Windows. Therefore,
installing the custom build alone is not enough: set backend to bridge in
the JSON config when full controller and touch support is required.
Requirements
Windows 10/11 for the patchless backend. The bridge backend follows the platform support of the custom Ryubing build.
Node.js 20 or newer.
A working Ryubing profile containing its
Config.json, keys, firmware, and any other data required to boot your games.The custom Ryubing MCP Bridge release for the complete tool set.
Installation
Option A: portable release layout
Download and extract the custom Ryubing build linked above.
Download the matching
v0.5.0MCP package from the GitHub releases page, or clone this repository and build it withnpm ci,npm run build, andnpm pack.Extract the generated
.tgz; its files are under thepackage/directory.Place the MCP package contents beside
Ryujinx.exeand install production dependencies withnpm install --omit=dev.Prepare the
portable/profile directory before creating a session.Change
backendtobridgeinryubing-mcp.config.json.
The resulting directory is self-contained:
ryubing-portable/
|-- Ryujinx.exe
|-- Ryujinx.dll and other Ryubing runtime files
|-- portable/ # ready profile template copied once per title
| |-- Config.json
| `-- system/ # prod.keys, title.keys, firmware/NAND data
|-- dist/
| `-- index.js
|-- node_modules/
|-- package.json
|-- ryubing-mcp.config.json
|-- ryubing-mcp.config.schema.json
`-- mcp-sessions/ # created automaticallyRun this once inside the directory:
npm install --omit=dev
node .\dist\index.jsThe second command starts an MCP stdio process and normally appears to wait; that is expected. MCP clients start and communicate with it automatically.
Option B: build from source
git clone https://github.com/Leuconoe/ryubing-mcp.git
Set-Location ryubing-mcp
npm install
npm run typecheck
npm test
npm run buildEither copy the custom Ryubing runtime/profile into this directory or keep them elsewhere and point an external JSON config at those paths.
MCP client registration
Zero-environment portable registration
When Ryujinx.exe, portable/, and the default config are beside dist/, no
environment variables are required:
{
"mcpServers": {
"ryubing": {
"command": "node",
"args": ["D:/Apps/ryubing-portable/dist/index.js"]
}
}
}Externally managed config
To keep runtime paths and settings outside the MCP package, inject only the config file path:
{
"mcpServers": {
"ryubing": {
"command": "node",
"args": ["D:/Apps/ryubing-mcp/dist/index.js"],
"env": {
"RYUBING_MCP_CONFIG": "D:/Config/ryubing-mcp.config.json"
}
}
}
}Use forward slashes or escaped backslashes in JSON paths. Restart the MCP client after changing its server registration or the Ryubing config.
Configuration file
The server loads configuration in this order:
The file named by
RYUBING_MCP_CONFIG.ryubing-mcp.config.jsonin the MCP process working directory.ryubing-mcp.config.jsonshipped beside the server package.
Relative portable.root paths are resolved from the directory containing the
selected config. Its three child paths are then resolved from portable.root.
Unknown settings and invalid values fail at startup instead of being ignored.
Complete bridge-oriented example:
{
"$schema": "./ryubing-mcp.config.schema.json",
"backend": "bridge",
"portable": {
"root": ".",
"executable": "Ryujinx.exe",
"profileTemplate": "portable",
"sessionRoot": "mcp-sessions"
},
"bridge": {
"host": "127.0.0.1",
"port": 49600,
"timeoutMs": 10000,
"launchTimeoutMs": 120000
},
"managed": {
"maxSessions": 4,
"portStart": 49610,
"startupTimeoutMs": 30000,
"retention": {
"keepLatest": 3,
"olderThanDays": 7,
"autoPrune": true
}
}
}JSON setting | Default | Description |
|
|
|
|
| Base directory for the following three portable paths |
|
| Custom or unmodified Ryubing apphost; a |
|
| Ready default profile copied for each title's first managed session |
|
| Parent directory for isolated session profiles |
|
| External/default bridge host; loopback values only |
|
| External/default bridge port |
|
| Timeout for each bridge request |
|
| Longer timeout for |
|
| Concurrent managed-session limit; maximum 32 |
|
| First port considered for managed bridge sessions |
|
| Time allowed for a newly launched emulator to become ready |
|
| Number of newest inactive unbound/exception profiles to retain; canonical title profiles are protected |
|
| Remove inactive profiles older than this many days; active processes are never removed |
|
| Run the safe inactive-profile cleanup before creating and after stopping a session; explicit cleanup remains available when disabled |
| omitted | Optional array of permitted game/update roots; omitted means unrestricted |
To restrict launchable files, add an allowlist. Relative entries are resolved from the config directory:
{
"allowedGameDirs": [
"games",
"E:/NSW/_titles/_waitng"
]
}allowedGameDirs is not required. Leave it out when arbitrary absolute game
paths must be usable. A configured allowlist rejects files outside its roots.
Optional environment overrides
Operational settings should normally remain in JSON. These variables exist for legacy launchers, CI, and secret injection, and override JSON when present:
Environment variable | Purpose |
| Select one external JSON config; normally the only injected variable |
| Override |
| Override external bridge host |
| Override external bridge port |
| Inject the external bridge bearer token without storing it in JSON |
| Override bridge request timeout |
| Override the operation-specific |
| Platform-delimited path allowlist override |
| Override the managed executable path |
| Override the profile-template path |
| Override the session-root path |
| Override the first managed bridge port |
| Override the managed-session limit |
| Override managed startup timeout |
Profile and session model
There are two kinds of session:
defaultrepresents an emulator bridge started outside this MCP server atbridge.host:bridge.port. OmittingsessionIdtargets it.A managed session is created by
ryubing_create_session. A title's first session copies the configured ready default profile, includingConfig.json,system/keys, and registered firmware/NAND. Later calls for that title reuse the running process or restart the retained profile instead of cloning again.Each managed profile contains
.ryubing-mcp-session.jsonand.ryubing-mcp/logs.ndjson. The manifest records lifecycle phase, last status, launch paths, exit code, and the last error. These files are retained when a process exits so diagnostics can distinguish a game failure from an MCP or bridge connection failure.Base, update, and child-program IDs belonging to one application share one title-session owner. For example,
...6000,...6001, and...6005all map to the...6000owner. Exact patch paths and runtime evidence must still use the actual child program ID selected by the emulator.
For normal automation, use managed sessions:
Call
ryubing_check_environmentand resolve everyerrorbefore a runtime test.Call
ryubing_validate_launchwith the base/update paths. Resolve everyerrorbefore starting a title.Call
ryubing_create_sessionwith the validatedapplicationId. Normally omitsessionId; the server selectstitle-<applicationId>and returns the existing title session when one is already active.Call
ryubing_launch_gamewith the returnedsessionIdand usewaitFor: "running"when the caller needs a readiness guarantee. A timeout is returned as retryable evidence by default; continue withryubing_wait_for_stateinstead of issuing a second launch.Use input, screenshot, frame-sequence, and log tools with the same
sessionId.If the game does not progress, call
ryubing_get_diagnosticsand, when sharing evidence,ryubing_export_diagnosticsbefore stopping.Call
ryubing_stop_sessionwhen finished and retain the profile for diagnosis.
Tool-level failures use a stable JSON envelope even when the MCP result has
isError: true:
{
"error": {
"code": "SESSION_TITLE_CONFLICT",
"message": "...",
"retryable": false
}
}Callers should inspect isError and error.code before consuming the result.
Timeout/readiness errors are marked retryable when the existing session can be
polled safely. The managed launcher also removes MCP configuration overrides
from the child emulator environment and injects only the per-session control
token and launch-specific variables.
ryubing_launch_game requires an explicit sessionId, including
sessionId: "default" when intentionally using an externally started bridge.
This prevents a caller that created a managed session from accidentally sending
the next title to the unrelated default emulator. Each normal managed session
is bound to one normalized application ID, and only one canonical managed
session may own that ID. Base/update paths for the same application ID can
change inside that retained session; a different title is rejected with
SESSION_TITLE_CONFLICT. Concurrent launches for one session are rejected with
SESSION_LAUNCH_IN_PROGRESS, so a timed-out request must be observed with
ryubing_wait_for_state rather than retried. Until a wait reaches a requested
state, another launch is rejected with SESSION_LAUNCH_OUTCOME_UNKNOWN, even
when it targets the same title.
The ready template is copied only for a title's first session, so the original
default emulator profile is never rewritten. A stopped canonical title profile
is automatically protected from retention and restarted in place; its copied
keys, firmware, saves, title keys, and settings therefore remain available.
Use removeProfile: true only when deliberately discarding that title profile.
Unbound sessions and reason-authorized parallel comparison sessions remain
subject to normal inactive retention so exceptions do not become another
unbounded storage source. Active processes and unrecognized directories are
never removed.
Patchless sessions also create a hard-linked runtime mirror, falling back to
normal copies when hard links are unavailable. This gives every process an
isolated Logs directory without normally duplicating the entire runtime.
Native PNG files used for patchless screenshots are temporary MCP evidence and
are pruned before and after each capture; returned image data remains available
to the MCP caller without building an unbounded screenshots/ archive.
MCP tool reference
Most emulator-control tools accept optional sessionId; omit it only when you
intentionally target an externally started default bridge. ryubing_launch_game
is the exception and requires an explicit ID to prevent title/session mix-ups.
ryubing_create_session
Starts a managed process and returns its session metadata and initial status.
For normal use, provide the 16-digit base applicationId and omit sessionId;
the stable ID is title-<applicationId>. Repeating the call returns the running
canonical session (reusedExisting=true) or restarts its retained profile
(reusedProfile=true). It does not recopy the template or require keys and
firmware to be injected again. reuseExisting remains only for legacy unbound
profiles.
Creating an unbound session or a deliberate second session for the same title
requires exceptionReason (10-500 characters), for example a simultaneous
baseline/patched comparison. The reason is persisted in the owned manifest.
Without it, duplicate creation fails with TITLE_SESSION_EXISTS and identifies
the session to reuse.
ryubing_list_sessions
Lists the external default entry plus all managed sessions, including backend,
state, PID, port where applicable, and profile directory.
ryubing_stop_session
Stops one managed process. sessionId is required. removeProfile defaults to
false; setting it to true recursively removes that isolated profile after
the process exits. Canonical title profiles are protected automatically so they
can be restarted without key/firmware setup. protectProfile: true remains
useful for unbound or exception sessions whose evidence must survive automatic
retention. The external default session cannot be stopped here.
ryubing_get_session_storage
Returns a read-only inventory of directories directly under
portable.sessionRoot, including ownership classification (active,
reclaimable, protected, legacy, or unconfirmed) and byte estimates.
Owned entries also expose applicationId and any duplicateTitleReason, making
pre-existing same-title profiles identifiable before explicit removal.
Directories without a matching MCP manifest are reported but never removed by
automatic cleanup.
ryubing_cleanup_sessions
Lists or removes inactive managed profiles under portable.sessionRoot.
dryRun defaults to true; set it to false only after reviewing the
matched session IDs, paths, and byte estimate. The default policy keeps the
newest three inactive profiles and removes profiles beyond that count or older
than seven days. Canonical title profiles are excluded; remove one explicitly
with ryubing_stop_session(removeProfile=true) when it is no longer needed. The
cleanup tool only accepts MCP manifests directly under the configured session
root, skips active PIDs, and never scans games, saves, keys, firmware, patches,
or arbitrary directories. It still works when managed.retention.autoPrune is
disabled.
Ryubing/Eden workspace cleanup skill
The repository includes
skills/nsw-workspace-cleanup for
shared title workspaces used by Ryubing MCP and Eden MCP. Cleanup is inferred
from the complete handoff chain, exact paths and hashes, title/project IDs,
ownership manifests, active PIDs, and pending review state. Names such as
tmp, copy, old, or a timestamp and duplicate hashes are never deletion
authority by themselves.
The bundled collector is read-only and rejects --apply/--delete. Supply the
title or repository roots, both emulators' relevant handoffs, and their session
roots when applicable:
node .\skills\nsw-workspace-cleanup\scripts\build-cleanup-context.mjs `
--root "E:\NSW\_titles\Title [0100000000000000]" `
--handoff "D:\workspace\____personal____\ryubing-mcp\HANDOFF.md" `
--handoff "D:\workspace\____personal____\eden-mcp\HANDOFF.md" `
--session-root "ryubing=E:\NSW\_tools\Ryubing\mcp-bridge-v0.1.0\sessions" `
--session-root "eden=E:\NSW\_tools\Eden\mcp-v0.4.0\sessions" `
--jsonUse the resulting snapshot to generate cleanup-plan.v1 with exact KEEP,
REVIEW, and ELIGIBLE_AFTER_APPROVAL records. A failed candidate can still be
the only evidence for an open diagnosis; an artifact referenced by either
emulator remains protected. If the inventory is truncated, a handoff is missing,
or title/session ownership conflicts, the plan must be BLOCKED. Actual cleanup
requires a separate user approval of exact instruction IDs and a fresh matching
snapshot. Prefer each emulator's MCP retention tools for owned sessions and
diagnostic bundles.
ryubing_get_status
Returns protocol/emulator version, process ID, state, title ID/name, and frame
dimensions where available. States are idle, loading, running, paused,
or stopping.
ryubing_get_capabilities
Returns the emulator-mcp-capabilities contract version 1.0 for the selected
session. It explicitly reports bridge-only analog/touch/multi-touch support,
available launch/capture/diagnostic operations, and current limitations. The
same default-session contract is available through the
emulator://ryubing/capabilities MCP resource for clients that discover
capabilities before selecting a tool.
ryubing_check_environment
Checks the managed executable, profile template, Config.json, standard
system/prod.keys and system/title.keys locations, registered firmware,
allowlisted game roots, session-root write access, and the selected session.
The result has overall: "ok", "warning", or "error" plus a checks array;
it does not read or return key contents. Set probeControl: true to query the
selected bridge status as part of the report. This is the first tool to call
when setup or firmware/key installation is uncertain.
Ryubing uses profile/system/ for keys with --root-data-dir. A legacy
profile/keys/ directory is reported as a warning and is not treated as a
successful standard key installation.
ryubing_press_buttons
Presses one or more buttons simultaneously and releases them after durationMs
(16-10000, default 100). player is 1-8 or handheld.
Supported button names:
A B X Y L R ZL ZR PLUS MINUS L_STICK R_STICK
DPAD_UP DPAD_DOWN DPAD_LEFT DPAD_RIGHT
SL_LEFT SR_LEFT SL_RIGHT SR_RIGHTPatchless mode accepts only Player 1/handheld and maps these names through the
WindowKeyboard bindings in the copied profile's Config.json.
ryubing_set_sticks
Sets any supplied leftX, leftY, rightX, or rightY coordinate in the
inclusive range -1 to 1. At least one axis is required. durationMs: 0 retains
the state; a positive duration up to 60000 resets both sticks afterward.
Requires the custom bridge build.
ryubing_touch
Sends tap, long_press, or swipe using Switch logical-screen coordinates:
x 0-1279 and y 0-719. A swipe also requires endX and endY.
durationMs is 16-10000. Requires the custom bridge build.
ryubing_take_screenshot
Returns metadata plus MCP image content. Bridge mode accepts png or jpeg
and JPEG quality 1-100. Patchless mode accepts PNG only, invokes Ryubing's
screenshot hotkey, and waits for the renderer PNG in the session profile.
Before bridge capture, the MCP checks control.status and requires non-zero
frameWidth/frameHeight; an early request returns retryable
SCREENSHOT_NOT_READY without calling the renderer capture path.
ryubing_get_logs
Reads up to 2000 entries (limit, default 200). Levels are trace, debug,
info, notice, warning, error, and critical. Pass the returned
nextCursor as the next call's cursor to receive newer entries only. The
optional contains value performs a case-insensitive match against each entry's
category and message and returns filter metadata while preserving the source
cursor.
ryubing_get_diagnostics
Collects a shareable JSON report without stopping or restarting the emulator.
It includes a reportVersion, capture time, selected session metadata, runtime
backend/platform, status, the requested trace-to-error log window,
cursor/truncation state, level counts, repeated-message first/last cursors,
and up to 100 warning/error/negative-log events with surrounding context. It
still returns logs when status collection fails; the errors object identifies
which part failed.
The focus argument prioritizes one of the common failure patterns:
| What it looks for |
| stuck |
| repeated patch/mod/LayeredFS/RomFS/ExeFS/retry messages |
| repeated video/movie/cutscene/NVDEC/codec/decoder messages |
| explicit missing, skipped, disabled, invalid, or failed patch messages |
| all four heuristic categories (default) |
Hints are evidence-based heuristics, not a proof of the root cause. Each hint
contains summary, bounded evidence, and recommendations; a focused call
with no matching signature returns a low-confidence “inconclusive” hint rather
than an empty result. Always pass the original logs.entries,
observations.events, and observations.repeatedMessages to the report when
escalating an issue. A useful capture is:
{
"sessionId": "qa-01",
"cursor": "0",
"limit": 2000,
"minimumLevel": "trace",
"focus": "all"
}Capture a fresh diagnostic after the failure, before launching another title or deleting the managed profile; the retained profile and cursor make repeated failures comparable.
When the managed process has already exited, the tool falls back to the
manifest's last status and the persisted NDJSON log. The report exposes
sources.status (live or manifest) and sources.logs (live or
persisted) and adds a warning instead of hiding the original failure behind
ECONNREFUSED.
ryubing_export_diagnostics
Collects the same report and writes report.json, the captured logs.ndjson,
and a SHA-256 manifest.json under the selected managed profile's
.ryubing-mcp/diagnostics/<timestamp>-<pid>/ directory. The result includes
the exact paths, byte count, and redaction count. Absolute paths and
secret-like fields or inline values are redacted by default. Export is
local-only; game contents, screenshots, keys, firmware, and environment
variables are not copied. Only set redact: false when the bundle remains on
the same trusted machine.
ryubing_get_runtime_metrics and ryubing_advise_compatibility
The metrics tool summarizes bounded FPS and frame-time samples plus shader and
video-failure signals found in structured logs. Missing counters remain null
or empty instead of being estimated. The advisor combines those metrics with
diagnostic hints and returns reversible, one-variable experiments with their
reason, tradeoff, and rollback. It never changes emulator settings.
ryubing_list_artifacts and ryubing_cleanup_artifacts
Artifact listing is limited to Ryubing MCP's exact diagnostic-directory naming
convention. Cleanup applies age and keep-latest retention to the same bounded
set and defaults to dryRun: true. It does not scan or delete games, managed
profiles, saves, patches, keys, firmware, or arbitrary user files.
ryubing_capture_sequence
Captures 2-12 frames at a bounded interval (50-10000 ms). The metadata includes
SHA-256 hashes, dimensions, optional bridge frame numbers, byte sizes, and
changedFromPrevious. It classifies exact-byte sequences as frozen,
periodic, or moving and reports the period and compared pairs. Image blocks
are included only while the total payload is under 24 MiB. Exact PNG/JPEG hashes
are strong evidence for identical output but less tolerant than Eden MCP's
grayscale perceptual comparison; compression differences can make equivalent
frames appear changed.
ryubing_run_input_script
Validates all operations before sending the first input, then executes up to 100
wait, buttons, sticks, touch, or wait_frame_change steps. The
declared wait/hold/timeout budget is limited to five minutes. Each result contains
the operation, duration, success, and result or failure. Stick players touched by
the script are reset to zero after failure and by default after completion.
{
"sessionId": "qa-01",
"steps": [
{"op":"buttons","buttons":["L","R"],"player":1,"durationMs":120},
{"op":"buttons","buttons":["A"],"player":1,"durationMs":80},
{"op":"wait_frame_change","timeoutMs":10000,"pollMs":500}
]
}Patchless sessions support wait, buttons, and screenshot-based frame
checkpoints. Stick and touch steps return the existing explicit Pending error
and the script preserves completed-step evidence.
ryubing_validate_launch
Performs the launch checks without starting Ryubing: path allowlist and
extension, file/directory metadata, filename title-ID hints, and base/update
compatibility. overall: "error" blocks ryubing_launch_game; warnings are
non-fatal but should be reviewed.
ryubing_inspect_patch
Scans an unpacked patch directory without modifying it. It reports title-ID
hints, romfs/exefs/LayeredFS markers, patch archive extensions, duplicate
relative paths, file count, and total bytes. This is a layout check only; it
does not decrypt NCA files or prove that a translation rendered in-game.
For the current Ryubing profile layout, install mods under
mods/contents/<program-id>/<mod-name>/romfs|exefs, or Atmosphere-compatible
content under sdcard/atmosphere/contents/<program-id>/. A legacy load/
directory is not searched by this build. In a multi-program title,
<program-id> is the exact child ID that consumes the patch, not necessarily
the base application ID used for session ownership.
ryubing_validate_font_atlas
Runs a read-only before/after check for a translated PNG font atlas. Supply the
original atlas, candidate atlas, a JSON glyph map, and the translated text
sample used for required glyph coverage. The map can be a single
glyphs array, separate reference/candidate arrays, or section objects with
glyphs and kernings. Each glyph identifies one Unicode code point and its
x, y, width, height, optional advance, baseline, and signed
offsetX/offsetY (xOffset/yOffset aliases are accepted). Kerning records
identify first, second, and signed amount values.
The harness fails on canvas or pixel-format changes, missing alpha, duplicate or overlapping rectangles, out-of-bounds glyphs, empty/clipped candidate glyphs, position/advance/baseline/offset/kerning drift, suspicious ink-size or occupancy changes, and missing glyphs in the supplied translated text sample. PNG and map sizes, decompression output, glyph count, rectangle work, and overlap comparisons are bounded to keep malformed inputs from exhausting the MCP process.
A clean first run is PENDING_USER_REVIEW and returns approval.reviewToken.
That token binds the reference PNG, candidate PNG, glyph-map JSON, translated
text, and tolerances. After the user approves the exact before/after evidence,
rerun with approvedReviewToken set to that token. Any artifact or parameter
change invalidates it with APPROVAL_TOKEN_MISMATCH; a boolean approval flag is
not accepted. This static PASS is still not a runtime or release PASS.
Example glyph map:
{
"reference": [{"codePoint":"U+AC00","x":2,"y":2,"width":32,"height":40,"advance":32,"baseline":30,"offsetY":-2}],
"candidate": [{"codePoint":"U+AC00","x":2,"y":2,"width":32,"height":40,"advance":32,"baseline":30,"offsetY":-2}],
"referenceKernings": [{"first":"U+AC00","second":"U+B098","amount":-1}],
"candidateKernings": [{"first":"U+AC00","second":"U+B098","amount":-1}]
}Keep failed candidates and the returned SHA-256 values in the title-local QA evidence. Fix the atlas in a new temporary staging path, rerun the harness, and request user review again; never overwrite the only reference/candidate pair.
ryubing_launch_game
sessionId is required. Use the returned ID from ryubing_create_session, or
use the literal default only when the external bridge is the intended target.
The server checks the bridge state before launching, serializes launches per
session, and verifies any reported title ID against the requested application.
Once a session has a launch identity, switching to another title is blocked.
A different base/update path with the same
normalized application ID stays in that title's existing session.
Launches an absolute .xci, .nsp, .nca, .nro, or unpacked game directory.
Optional updatePath must be a matching update .nsp. The update applies only
to the isolated session. Supply the 16-digit hexadecimal applicationId when
the update filename does not expose a title ID.
For a user-owned encrypted dump that requires an external title key, either call
ryubing_inject_title_key first or pass the same object as titleKey on
ryubing_launch_game:
{
"sessionId": "qa-01",
"basePath": "E:/Games/Example.nsp",
"titleKey": {
"rightsId": "00112233445566778899AABBCCDDEEFF",
"key": "FFEEDDCCBBAA99887766554433221100"
}
}Both values must be exactly 32 hexadecimal characters. Injection is limited to
an idle isolated managed session and atomically merges into that session's
system/title.keys; the external default profile and profile template are
never changed. The key itself is not returned. Repeating the same pair is a
no-op. A different key for an existing Rights ID fails with
TITLE_KEY_CONFLICT unless overwrite: true is included. Ryubing reloads the
session key set as the title is initialized, so launch-time injection does not
require restarting the managed process.
waitFor defaults to "accepted" for compatibility. Set it to "running" to
poll the selected bridge until control.status.state is running; the bounded
waitTimeoutMs (default managed.startupTimeoutMs) returns
targetReached=false, timedOut=true, and retryable=true without stopping
the selected process. Set keepRunningOnTimeout=false only when legacy tool
error behavior is required. bridge.launchTimeoutMs is a separate, longer
timeout for the launch RPC itself.
If that RPC deadline expires before the bridge acknowledges the launch, the
default result is also non-terminal: launch.accepted is "unknown",
launch.requestTimedOut=true, and retryable=true. This does not prove that the
emulator rejected the game. Keep the owned process, poll
ryubing_wait_for_state, and collect diagnostics from that session. Do not
issue a second game.launch while the first request's outcome is unknown;
SESSION_LAUNCH_OUTCOME_UNKNOWN remains active until a state wait reaches its
target or the session is stopped/recreated.
ryubing_wait_for_state
Continues polling the existing selected session for idle, loading,
running, paused, or stopping without sending another game.launch. Use it
after a retryable launch timeout. The response includes the last status/error,
elapsed time, and whether another wait remains safe.
Example call arguments:
{
"sessionId": "qa-01",
"basePath": "E:/NSW/_titles/_waitng/Example [0100000000000000].xci",
"updatePath": "E:/NSW/_titles/_waitng/Example Update [0100000000000800].nsp",
"applicationId": "0100000000000000"
}Full bridge security model
The bridge binds only to
127.0.0.1,::1, orlocalhost; remote network control is deliberately unsupported.Managed bridge sessions choose separate ports and generate a cryptographically random token for each process.
On Windows the token is passed through a per-session token file rather than a visible command-line argument.
The MCP server validates protocol version
0.1, input ranges, paths, response IDs, screenshot encoding, and screenshot size.Keep MCP access local and do not expose stdio or bridge ports through an unauthenticated network proxy.
Patchless behavior and limitations
Patchless mode is a compatibility implementation, not the complete control protocol. It:
Detects PID, emulator version, title name, and title ID from the process and Ryubing window title.
Delivers configured digital keyboard keys directly to the Ryubing window.
Uses Ryubing's configured screenshot hotkey and returns the new PNG.
Parses the latest isolated runtime log using a numeric cursor.
Writes a session-local
games/<titleId>/updates.jsonwhen an update is used, then restarts Ryubing with the base game path.Returns a clear
Pendingerror for analog-stick and touch requests.
Because it relies on native window/input behavior, keyboard bindings must exist
and security or overlay software may interfere. Switch to backend: "bridge"
before treating input automation as release-quality evidence.
Troubleshooting
First response for any abnormal gameplay
Do not immediately restart, delete the profile, or apply another patch. Preserve the failing session so its cursor and profile remain useful:
Call
ryubing_check_environment({"sessionId":"qa-01","probeControl":true}).Call
ryubing_get_status({"sessionId":"qa-01"}).Call
ryubing_get_diagnosticswithcursor: "0",limit: 2000,minimumLevel: "trace", and the most relevantfocusvalue below.Save the complete JSON response, emulator build, firmware version, title ID, base/update paths, session ID, and the fresh screenshot separately.
Only then restart the title or try one controlled change.
The diagnostic hints are heuristics. A loader message, a running process, or a
successful patch copy alone does not prove that the game reached the expected
screen or that Korean data was rendered.
Game does not load or remains on loading
Run ryubing_validate_launch first and resolve any title-ID mismatch. Then use
focus: "game_load_failure". Check the environment report first: the
executable, profile, system/prod.keys, system/title.keys, and registered
firmware must be present. Confirm the base path exists and that an update is
matched to the base title. If the update filename does not contain the base
title ID, pass applicationId explicitly. Compare status.state, titleId,
and the last load/guest/error entries in the diagnostic report.
Wrong patch causes a loop or prevents progression
Run ryubing_inspect_patch on each candidate and keep its output with the
report. Then use focus: "patch_loop". Inspect observations.repeatedMessages and the
patch_loop evidence rather than only the final error. Keep exactly one enabled
patch candidate for the selected title root, verify the consumer title ID, and
restart the title after changing patch files. A retained managed profile should
be used for the before/after comparison; do not mix a default profile log with a
managed-session result.
Video or cutscene repeats instead of progressing
Use ryubing_capture_sequence first, then focus: "video_loop" and the complete
diagnostic log window. Look for repeated NVDEC, codec, decoder, movie, cutscene,
media, or frame messages. Separate a video decoder failure from a patch loop:
the process can remain running while the render surface is not advancing.
Record the title ID, firmware, graphics backend, and whether the same scene
repeats after a clean unpatched launch.
Patch exists but Korean data is not displayed
Use focus: "patch_not_applied". Verify that the patch is staged under the
actual consumer title ID, that only the intended candidate is enabled, and that
the managed profile is the profile being launched. Restart the title after
staging; LayeredFS/ExeFS or file-copy messages are not proof that the selected
translation reached the rendered text. Compare a baseline capture with a fresh
patched capture and retain both diagnostic reports.
Keys or firmware were installed manually but the game still fails
For a --root-data-dir profile, Ryubing reads key files from
<profile>/system/, not <profile>/keys/. Use the environment report to detect
the legacy location. Supply a real firmware ZIP containing NCA entries; an
incomplete archive or an unrelated executable ZIP will fail before installation.
New titles receive the corrected ready template automatically. An existing
canonical title session intentionally retains its own working copy to avoid
repeated key/firmware setup; correct that profile in place, or explicitly stop
it with removeProfile: true and create it again with the same applicationId
when a full reset is intended.
Ryubing MCP config file not found
Check RYUBING_MCP_CONFIG, use an absolute path, and remember that JSON paths
need forward slashes or escaped backslashes.
Managed executable or profile does not exist
Confirm portable.root, portable.executable, and
portable.profileTemplate. The executable must be a file and the template must
already be a usable directory. sessionRoot is created automatically.
Bridge startup/readiness timeout
Verify that the executable is the custom bridge release and that config uses
backend: "bridge". Increase managed.startupTimeoutMs for slow firmware/game
startup. The managed session searches available loopback ports beginning at
managed.portStart.
Pending: analog stick or Pending: touchscreen
The session is using patchless mode. Install the custom build and explicitly set
backend to bridge.
Patchless button is unbound or unsupported
Open the template profile in Ryubing and configure Player 1 WindowKeyboard
bindings so new titles inherit them. For an existing canonical title session,
either update its copied Config.json or explicitly remove and recreate that
title profile.
Patchless screenshot times out
Confirm the screenshot hotkey in Config.json is bound and that Ryubing can
write PNG files under the session profile's screenshots/ directory.
Game path is rejected
Use an absolute supported path. If allowedGameDirs is present, the real file
must be inside one of those directories. Remove the setting to allow all paths.
Update title ID cannot be inferred
Pass the base game's 16-digit applicationId explicitly. Do not pass the update
title ID ending in the update suffix.
Old session directory already exists
Call ryubing_create_session with the same applicationId; a canonical title
profile is restarted in place automatically. reuseExisting: true is only for
legacy unbound profiles. If an intentional reset is required, remove the
previous session through ryubing_stop_session with removeProfile: true and
then create it again. Directories without a valid MCP manifest are never
deleted automatically.
Emulator exits without an error log
When a managed process exits with exitCode: null and the persisted log has no
error or critical entry, diagnostics label the result as an inconclusive
runtime anomaly. A briefly reported child program, connection refusal after
exit, or absence of patch-loader records does not prove that the title or patch
failed. Preserve the session, collect diagnostics, and require an exact child
program plus PatchExeFS/PatchRomFS evidence before assigning a patch verdict.
Verification and development
Run the local quality checks:
npm run typecheck
npm test
npm run buildWith an external bridge already running at the configured default endpoint:
npm run smoke:liveFor two managed patchless processes, configure the managed paths and set
RYUBING_LIVE_BASE_A, RYUBING_LIVE_UPDATE_A, and RYUBING_LIVE_BASE_B before
running:
npm run smoke:parallelThe smoke tests require user-owned games and are not part of the normal unit test run.
For a read-only preflight matrix over a directory such as
E:/NSW/_titles/_waitng, set the directory outside the MCP process config and
run:
$env:RYUBING_GAME_MATRIX_DIR = 'E:\NSW\_titles\_waitng'
npm run qa:matrixThe matrix reports one ryubing_validate_launch result per .xci, .nsp,
.nca, or .nro file. It does not launch games or copy/modify their files;
use it before a controlled runtime matrix with one retained managed profile per
title.
Releases and source
MCP server releases: https://github.com/Leuconoe/ryubing-mcp/releases
Custom Ryubing source fork: https://github.com/Leuconoe/Ryubing
Custom Windows build: https://github.com/Leuconoe/Ryubing/releases/tag/mcp-bridge-v0.2.0
License
This MCP server is licensed under the MIT License. Ryubing and its dependencies retain their respective licenses.
Available Tools
10 toolsryubing_create_sessionCreate isolated Ryubing sessionA
Start a new Ryubing process with an isolated profile, bridge port, and authentication token.
| Name | Required | Description | Default |
|---|---|---|---|
| sessionId | No | Stable session ID; omit to generate one |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate a mutating operation (readOnlyHint=false), and the description adds meaningful context about what the session includes (isolated profile, bridge port, authentication token). It does not mention process lifecycle, resource usage, or that sessions should be stopped later, but the key side effect of starting a process is disclosed.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single, concise sentence that is front-loaded with the action ('Start a new Ryubing process') and adds essential details without redundancy. Every word earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a low-complexity tool with one optional parameter and no output schema, the description covers the core purpose and key behavioral elements. It does not mention return values or explicitly tie to stop_session, but the overall context from annotations and the simple schema makes it reasonably complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema fully describes the single parameter (sessionId) with pattern and description, so baseline is 3. The tool description does not add extra parameter semantics beyond what the schema already provides.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb ('Start') and resource ('Ryubing process') with distinct attributes ('isolated profile, bridge port, authentication token'), clearly differentiating it from siblings like list_sessions and stop_session. The title confirms the creation nature.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is implied by the tool name and description (creates a session before other actions), but there is no explicit guidance on when to use it vs alternatives, prerequisites, or exclusions. Sibling tools suggest lifecycle but the description itself lacks direct instructions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_get_logsGet Ryubing logsARead-only
Read structured emulator logs. Pass the returned cursor to get only newer entries on the next call.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | ||
| cursor | No | ||
| sessionId | No | Managed Ryubing session ID; omit for the external default session | |
| minimumLevel | No | info |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and destructiveHint=false, establishing the safe read-only nature. The description goes beyond annotations by disclosing the cursor-based incremental pagination behavior, which is not evident from schema or annotations. This adds meaningful behavioral context without contradiction.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description consists of two concise front-loaded sentences. The first clearly states the core purpose, and the second provides a crucial operational detail without any redundancy or fluff. Every word earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with 4 optional parameters and no output schema, the description is adequate but minimal. It covers the key pagination concept and benefits from strong read-only annotations. However, it does not mention filtering by minimum level, default limits, or what the response structure looks like beyond implying a cursor, leaving some gaps for an agent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is low (25%, only sessionId has a description). The description adds meaning to the cursor parameter by explaining its role in fetching newer entries. However, limit and minimumLevel are not elaborated in the description, though their names and schema defaults provide some intrinsic semantics.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description 'Read structured emulator logs' uses a specific verb ('read') and resource ('structured emulator logs'), clearly distinguishing it from sibling tools that control the emulator or manage sessions. Even though the title says 'Get logs', the description adds 'structured' and clarifies the read-only nature, making the purpose unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage when logs need to be retrieved, but does not explicitly discuss alternatives or when not to use the tool. However, the cursor instruction ('Pass the returned cursor to get only newer entries on the next call') provides a clear usage tip for incremental reading, which adds some guidance beyond mere implication.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_get_statusGet Ryubing statusARead-onlyIdempotent
Check bridge compatibility and the currently running game.
| Name | Required | Description | Default |
|---|---|---|---|
| sessionId | No | Managed Ryubing session ID; omit for the external default session |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and idempotentHint=true, so the agent knows this is a safe read operation. The description adds the specific items being checked but does not disclose response format or potential edge cases. This is adequate for a simple status tool but not rich behavioral context.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence, front-loaded with the action and outcome. Every word earns its place with no redundancy or fluff.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with one optional parameter and no output schema, the description names the two key outputs (bridge compatibility and running game). However, it does not elaborate on what 'bridge compatibility' means or the response structure, which would be helpful given no output schema. Overall, it is fairly complete for the tool's simplicity.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema fully describes the single optional parameter sessionId with a pattern and explanation, giving 100% schema coverage. The description adds no additional parameter details, so the baseline score of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's function with a specific verb 'Check' and identifies exactly what it reports: bridge compatibility and the currently running game. This distinguishes it from sibling tools like launch_game, stop_session, and take_screenshot, which perform different actions.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies use when you need to verify status or compatibility, but it provides no explicit guidance on when to prefer this over other tools like get_logs or list_sessions. No alternatives or exclusions are mentioned, so usage context is only implicit.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_launch_gameLaunch a Switch gameA
Launch a user-owned base game file, optionally applying a matching update for this session only.
| Name | Required | Description | Default |
|---|---|---|---|
| basePath | Yes | Absolute path to .xci, .nsp, .nca, .nro, or an unpacked game directory | |
| sessionId | No | Managed Ryubing session ID; omit for the external default session | |
| updatePath | No | Absolute path to a matching update .nsp |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations are all false, indicating a mutating, non-idempotent, non-destructive operation. The description adds the behavioral detail that updates apply only for the session, which is useful. However, it does not mention potential side effects like starting or stopping the emulator, or any state changes beyond the session update.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence, front-loaded with the verb 'Launch,' and conveys the essential purpose and optional behavior with zero waste.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple launch tool with three well-documented parameters and no output schema, the description plus schema provides sufficient context. It could have mentioned what the return value or success/failure indication looks like, but that is not strictly necessary for invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, with all three parameters already well-described (e.g., basePath formats, sessionId format, updatePath purpose). The description adds no new parameter semantics beyond the schema, so the baseline of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Launch a user-owned base game file' with a specific verb and resource, and mentions the optional update application. It is immediately distinguishable from sibling tools like ryubing_get_status or ryubing_take_screenshot.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies clear usage: to start a game, possibly with an update. It does not explicitly name alternatives or exclusions, but the context of launching a game is unambiguous given the sibling set. The mention of 'for this session only' also provides a usage constraint.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_list_sessionsList Ryubing sessionsARead-onlyIdempotent
List the external default bridge and all managed Ryubing processes.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, and destructiveHint=false, covering the safety profile. The description adds specificity about the list's contents, but does not disclose any additional behavioral traits like return format or whether authentication is needed. This matches the baseline where annotations do the heavy lifting.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single focused sentence with no filler or repetition. It efficiently conveys the tool's purpose without wasting words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
This is a simple list operation with no parameters and no output schema. The description sufficiently indicates what will be listed, though it does not explicitly mention the return format. Given the tool's simplicity, this is adequate.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The tool has zero parameters, and schema coverage is trivially 100%. The baseline for no parameters is 4, and the description correctly does not attempt to describe non-existent parameters.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses the specific verb 'List' and names the exact resources ('external default bridge and all managed Ryubing processes'), making the purpose unmistakable and distinguishing it from sibling tools like get_status or create_session.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is provided on when to use this tool rather than alternatives such as ryubing_get_status or ryubing_stop_session. The description states only what the tool does, not the context or conditions for its use.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_press_buttonsPress Switch buttonsA
Press one or more Nintendo Switch controller buttons simultaneously, then release them after a duration.
| Name | Required | Description | Default |
|---|---|---|---|
| player | No | Controller player number or handheld mode | |
| buttons | Yes | Buttons to press together | |
| sessionId | No | Managed Ryubing session ID; omit for the external default session | |
| durationMs | No |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses the key behavioral trait: buttons are pressed simultaneously and held for a specified duration before release. Annotations provide no safety hints (all false), so the description carries the burden, and it does so well for a simple input tool. It does not mention side effects like blocking or output, but these are not critical.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that front-loads the action and outcome. Every word is necessary and there is no fluff.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple action tool, the description covers the core behavior and timing. The schema handles parameter details. It lacks an explicit statement about return values, but no output schema exists, and the action is straightforward enough that this is not a significant gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 75% (3 of 4 params have descriptions). The description itself does not add detail about parameters beyond what the schema provides, and durationMs is left without a description. The baseline for high schema coverage is 3, and the description does not compensate for the missing parameter description.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's function: pressing one or more Nintendo Switch controller buttons simultaneously and releasing after a duration. It uses a specific verb ('Press') and resource ('Switch controller buttons'), and is distinct from sibling tools like ryubing_set_sticks or ryubing_touch.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description clearly implies the tool is for button inputs, and no exclusions are stated. However, it does not explicitly mention alternatives or when-not-to-use, so it falls short of a 5. The context is clear: use this when you need digital button presses.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_set_sticksSet analog sticksA
Set left and/or right analog stick coordinates. Values range from -1 to 1. Omitted sticks keep their prior state.
| Name | Required | Description | Default |
|---|---|---|---|
| leftX | No | ||
| leftY | No | ||
| player | No | ||
| rightX | No | ||
| rightY | No | ||
| sessionId | No | Managed Ryubing session ID; omit for the external default session | |
| durationMs | No | Reset both sticks after this delay; 0 keeps the state until the next call |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With annotations all set to false, the description carries the burden of behavioral disclosure. It reveals the stateful nature of sticks (omitted sticks retain their previous state), which is valuable beyond the annotations. It does not address idempotency or side effects, but the 'set' action is clear.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is one concise, front-loaded sentence that conveys the essential purpose and a key behavioral note. No wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool has 7 parameters and no output schema, the description is minimal. It covers the core stick-setting behavior but omits important context like player targeting and the durationMs reset mechanism, which remain only in the schema. This is adequate but not fully comprehensive.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is only 29% (2 of 7 params documented). The description adds range and omission semantics for the stick coordinates, partially compensating for the low coverage. However, it does not explain the 'player' parameter or the durationMs reset behavior, though durationMs is described in the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states the tool's function with a specific verb and resource: 'Set left and/or right analog stick coordinates.' This clearly distinguishes it from sibling tools like press_buttons and touch, which handle different input types.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives a practical usage hint ('Omitted sticks keep their prior state'), but it does not explicitly say when to use this tool instead of alternatives. No exclusions or alternative tool names are mentioned, so the agent must infer tool selection from the name and context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_stop_sessionStop managed Ryubing sessionADestructive
Stop one managed Ryubing process. Its isolated profile is retained unless explicitly removed.
| Name | Required | Description | Default |
|---|---|---|---|
| sessionId | Yes | ||
| removeProfile | No |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description adds behavioral context beyond annotations by clarifying that the process is stopped individually and the profile is retained by default, unless explicitly removed. This complements the destructiveHint=true annotation without contradicting it.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two concise sentences. The first states the primary action, and the second adds a crucial caveat about profile retention. No unnecessary words, and it is front-loaded with the main purpose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple two-parameter destructive tool, the description is fairly complete. It explains the core action and a key side effect. It lacks details about error conditions or post-stop state, but given the simplicity and annotations, this is acceptable.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With 0% schema description coverage, the description must compensate. It partially does by hinting at the 'removeProfile' parameter ('unless explicitly removed'), but it does not add details about 'sessionId' or clarify parameter syntax. The meaning of sessionId is obvious from the name, but the description could be more explicit.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose with a specific verb ('Stop') and resource ('one managed Ryubing process'). It distinguishes this from sibling tools like ryubing_create_session or ryubing_launch_game by focusing solely on termination.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear context for when to use this tool: to stop a managed session. It also implies usage for the 'removeProfile' behavior by noting the profile is retained unless explicitly removed, but it does not explicitly mention alternatives or when-not-to-use scenarios.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_take_screenshotTake a Ryubing screenshotARead-only
Capture the emulator's rendered game frame without desktop window chrome.
| Name | Required | Description | Default |
|---|---|---|---|
| format | No | png | |
| quality | No | Used only for JPEG | |
| sessionId | No | Managed Ryubing session ID; omit for the external default session |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and destructiveHint=false, covering the safety profile. The description adds the 'without desktop window chrome' scoping but does not disclose how the screenshot is returned (e.g., file path, base64) or any side effects. With annotations covering safety, this is acceptable but not rich.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single, front-loaded sentence with no waste. 'Capture the emulator's rendered game frame without desktop window chrome' is specific and succinct, earning every word.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description defines what is captured but omits details about the return value or output format, which is critical since no output schema exists. It also lacks explicit usage guidelines. Given the simple optional parameters and strong annotations, it is minimally adequate but has a clear gap on return behavior.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema covers 67% of parameters with descriptions (quality, sessionId), while format lacks a description. The tool description adds no parameter-specific context, relying on the schema. At the high schema coverage baseline, this is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses the specific verb 'Capture' with the resource 'the emulator's rendered game frame' and adds the scope 'without desktop window chrome,' clearly distinguishing it from less specific screenshot tools. There are no competing siblings, so the purpose is unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for capturing pure gameplay visuals ('without desktop window chrome') but does not explicitly state when to use it, when not to, or mention alternatives. Sibling tools do not include another screenshot tool, so the guidance is minimal but not misleading.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
ryubing_touchTouch the Switch screenA
Send a tap, long press, or swipe using native Switch touchscreen coordinates (1280x720).
| Name | Required | Description | Default |
|---|---|---|---|
| x | Yes | ||
| y | Yes | ||
| endX | No | ||
| endY | No | ||
| action | Yes | ||
| sessionId | No | Managed Ryubing session ID; omit for the external default session | |
| durationMs | No |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations are all false, offering no meaningful behavioral signal. The description adds useful context about the coordinate system (1280x720) and action types, but does not disclose whether the tool blocks until completion, requires specific permissions, or has side effects beyond the input. This leaves some ambiguity.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, concise sentence that front-loads the essential information (action types and coordinate system) without redundancy or extra filler. Every word earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With 7 parameters, no output schema, and sparse annotations, the description is adequate but not comprehensive. It gives essential orientation but leaves questions about swipe endpoints, duration semantics, and sessionId usage. The schema fills some gaps, but the overall package is not rich.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is only 14%, so the description must compensate. It clarifies that x/y are native Switch coordinates and that action can be tap, long_press, or swipe, which helps interpret the core parameters. However, it does not explicitly explain endX/endY or durationMs, relying on schema defaults and inference.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool sends touch inputs (tap, long press, swipe) to the Switch screen using native coordinates. This specific verb+resource combination distinguishes it from sibling tools like ryubing_press_buttons and ryubing_set_sticks, which handle other input types.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies the tool is for touchscreen interactions, providing clear context for when it applies (e.g., tapping UI elements) versus other input tools. However, it does not explicitly mention alternatives or exclusions, so it falls just short of full explicit guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
10 tool updates
v0.1.0- First observed
ryubing_create_session - First observed
ryubing_get_logs - First observed
ryubing_get_status - First observed
ryubing_launch_game - First observed
ryubing_list_sessions - First observed
ryubing_press_buttons - First observed
ryubing_set_sticks - First observed
ryubing_stop_session - First observed
ryubing_take_screenshot - First observed
ryubing_touch
TDQS
Scored across 10 tools
Each tool targets a distinct resource or action: session management, status, game launch, input methods, screenshot, and logs. There is no overlap, and descriptions clearly distinguish between session status and active game status.
All tools use the 'ryubing_' prefix and mostly follow a verb_noun pattern (e.g., press_buttons, set_sticks, launch_game, list_sessions). The only deviation is ryubing_touch, which is a single verb, but it remains clear and does not cause confusion.
Ten tools is a well-scoped size for an emulator control server, covering session lifecycle, input, diagnostics, and game launching without redundancy. Each tool earns its place.
The tool set covers the core workflow: create session, launch game, control input, take screenshots, read logs, and stop session. Missing advanced features like save states or configuration settings are minor gaps for the apparent purpose.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
The official MCP Server for the Mux API
An MCP server that let you interact with Cycloid.io Internal Development Portal and Platform
MCP server to assist with JxBrowser development.
Related MCP Servers
- AlicenseAqualityBmaintenanceMCP server for the mGBA Game Boy Advance emulator. Read and write GBA memory, inject button presses, take screenshots, save/load state, and step the emulator through a Lua bridge.18102MIT
- AlicenseCqualityDmaintenanceAn MCP server for real-device Android reversing workflows, wrapping AlgorithmAide config writes, AppSwitch/logList sync, LSPosed scope sync, Frida script injection, and runtime log queries into a stable MCP toolset.2313MIT
- AlicenseNot gradedqualityCmaintenanceLocal MCP server for Android and iOS mobile automation and performance telemetry, enabling device control (screenshot, tap, swipe, input, app launch) and metric collection (CPU, memory, launch time) via ADB, simctl, and WebDriverAgent with SQLite session history.131MIT
- AlicenseNot gradedqualityCmaintenanceMCP server for controlling Hyprland Wayland compositor: screen capture, mouse/keyboard injection, and automation via native Hyprland IPC and wlr protocols.MIT