JARVIS-MCP
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., "@JARVIS-MCPcheck git status and recent commits in my active project"
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.
What is jarvis?
jarvis is the current Windows-first local development gateway used to bootstrap the next-generation JARVIS MCP development track.
Related MCP server: mcp-opentym
Current version: v4.13.0
The current v4.30 merge track contains 239 configured tools in the complete runtime inventory, with 194 advertised by default. The remaining inventory is explicitly filtered by delivery state (feature_disabled, planned, or dependency-gated) rather than being advertised as operational. The earlier 184-tool snapshot remains only as the compatibility baseline used by the v4 architecture; newer gateway capabilities are additive.
Current Windows 10/11 x64 artifacts are jarvis-Setup-4.13.0.exe (recommended installer) and jarvis-Portable-4.13.0.exe (no installation required), produced by the packaging pipeline at:
apps/desktop/dist/installers/jarvis-Setup-4.13.0.exe
apps/desktop/dist/installers/jarvis-Portable-4.13.0.exeRepresentative tool-permission policy rows (Safe asks, Balanced/Full allow):
Tool | Permission | Notes |
workspace_list | EXECUTE | Read-only discovery of registered roots and workspaces. |
JARVIS MCP development track
JARVIS MCP is the product direction for a project-aware AI development gateway optimized for ChatGPT Web. It is a Node.js/TypeScript MCP server with a local Web UI, not an Electron/Windows EXE.
The active JARVIS repository is JARVIS-MCP at C:\Users\JOJO\JARVIS-MCP. Start the complete local JARVIS runtime with start.bat from the repository root.
Web UI:
http://127.0.0.1:18767/MCP endpoint:
http://127.0.0.1:18766/mcpRuntime state:
%LOCALAPPDATA%\\jarvis-mcp
The Web UI currently provides:
Dashboard — runtime health, tool count, active project, Git/process state, knowledge status, and endpoints.
Projects — register local development projects and select the active project; the selection is also the live MCP active-workspace scope used by project-aware tools and Dashboard/workspace snapshots.
Live Log — inspect a compact recent diagnostics/activity view, toggle DEBUG visibility, and clear the backing log immediately.
Settings — configure theme, refresh cadence, Obsidian vault, and log display size.
JARVIS writes its UTF-8 diagnostics log to the active workspace root (log.txt) by default and clears that file at every JARVIS startup so each run has a clean diagnostic session. The diagnostics path can be overridden with JARVIS_DIAGNOSTICS_PATH for isolated/test runtimes; the Live Log Clear log action clears the configured backing file, and the assistant can read it directly from the active workspace when diagnosing failures.
JARVIS currently reuses the proven jarvis runtime/service packages while it is being developed. This is a temporary implementation substrate, not the target product boundary. JARVIS has its own server identity, ports, runtime data directory, Web UI, and product-specific knowledge tools.
The intended end state is for JARVIS to own its MCP runtime, Secure Tunnel, Web UI, project registry, Obsidian knowledge layer, automated development workflow, diagnostics, and self-development lifecycle. The legacy jarvis Desktop application is not part of the JARVIS product UI.
Start / Stop lifecycle
start.batderivesJARVIS_WORKSPACEfrom its own location, so the repository can be moved/renamed without changing a hard-coded workspace path.start.batrefuses to launch a second JARVIS instance when ports18766or18767are already listening.stop.batstops the JARVISnode dist/main.jsprocess tree and cleans up an orphaned tunnel listener on18768.If a terminal window is closed unexpectedly, run
stop.batbefore starting JARVIS again; this avoids leaving atunnel-client.exechild process behind.
OpenAI Secure MCP Tunnel
JARVIS owns its own Secure MCP Tunnel rather than sharing the legacy jarvis tunnel. The Web UI Settings page provides runtime configuration for the control-plane API key, tunnel_id, control-plane URL, auto-reconnect, and reconnect delay. The API key is protected with Windows DPAPI and is never returned by the Web API.
JARVIS prepares its own official OpenAI tunnel-client under %LOCALAPPDATA%\\jarvis-mcp\\tunnel\\bin. Configuration changes regenerate the JARVIS tunnel profile and the runtime manager performs doctor → run → health/poll verification → reconnect automatically. JARVIS reports the tunnel as connected only after the tunnel-client health check confirms a successful control-plane poll. The tunnel's main MCP binding points to JARVIS's own 127.0.0.1:18766/mcp endpoint.
Memory layer
JARVIS uses SQLite as the canonical durable memory store with Working, Episodic, Semantic, and Relational/Graph layers. Memory retrieval combines scoped FTS5 lexical search, optional local embedding similarity, and graph expansion. Global and project memory are isolated by default. Expired records and exact duplicates can be maintained deterministically without replacing the canonical SQLite source of truth; SQLite backup/restore and retention remain recoverable through the existing backup service.
Agent Harness Core
Phase 1 adds a durable Agent Harness above the existing Goal Continuation and Memory services. Agent Runs are persisted in SQLite (agent_runs and agent_run_events) with optimistic revision control and deterministic transitions:
created → understanding → contextualizing → planning → executing → observing → verifying → recovering → completed/failed/cancelled
The MCP surface provides jarvis_agent_run_create, jarvis_agent_run_status, jarvis_agent_run_transition, jarvis_agent_run_checkpoint, and jarvis_agent_run_control. These five tools are opt-in via JARVIS_AGENT_HARNESS_TOOLS=1 so the historical default MCP catalog remains byte-stable for existing connectors. Checkpoints preserve the current step, plan, working context, and result; runs can be resumed after process interruption. State transitions write working-memory records and auditable events. Capability execution remains delegated to the existing MCP/runtime layer so policy and Active Project boundaries are unchanged.
Intelligent Tool Router / Context Intelligence 2.0 / Verification Engine
Phase 2–4 add three additive JARVIS controls: route_intent deterministically ranks existing tools without bypassing ToolRegistry policy; context_packet produces a versioned 2.0 envelope combining project metadata, Git state, ranked workspace context, memory, and goal context; and verify_run executes the authoritative typecheck → tests → build → runtime-health → Git-diff evidence gates. Verification states are VERIFIED, PARTIALLY_VERIFIED, FAILED, or BLOCKED. Agent Harness completion is fail-closed: a run cannot transition to completed unless VERIFIED evidence is supplied.
Recovery Engine 2.0
Phase 5 adds deterministic failure classification and a fail-closed recovery policy. self_heal_plan now consumes recovery evidence through RecoveryEngine, binds plans to checkpoint/provenance metadata, and self_heal_apply revalidates the plan before any mutation. Verification/runtime/process/tunnel failures are classified separately, Git-change conflicts enter a human-review path, and recovery actions explicitly require review/confirmation where they can alter state. Automatic destructive retries are prohibited. Recovery mutations remain routed through the existing capability policy and carry provenance in their result/audit payload.
Self-development / Knowledge / Control Plane / Production readiness
The Phase 7–10 foundation keeps JARVIS on the same fail-closed development path it uses for normal projects. Self-development requires a checkpoint and authoritative verification evidence (typecheck, test, build, runtime, and git-diff) before completion can be reported as VERIFIED.
The Obsidian knowledge layer now derives a rebuildable metadata index from Markdown domains, tags, [[wikilink]] relations, SHA-256 content fingerprints, and filesystem modification times. The Markdown vault remains authoritative; the derived index is not.
The local Web control plane exposes /api/control-plane for MCP/tunnel/Active Project/index-watcher/memory/Agent Run state and /api/production-readiness for SQLite integrity, migration, path, and UTF-8 readiness checks. The Dashboard surfaces the control-plane and production-gate state alongside the existing runtime, tunnel, project, knowledge, and Git views.
JARVIS reads .obsidian-wiki by default and exposes read-only context tools for status, ranked search, note retrieval, backlinks, and compact project-development context. Set JARVIS_OBSIDIAN_PATH to use another vault.
Development model
The long-term JARVIS workflow is:
ChatGPT Web → JARVIS Secure Tunnel → JARVIS MCP → project context → inspect → edit → verify → summarize
JARVIS is itself a development target. The same project-aware capabilities used to improve other repositories will ultimately be used to inspect, modify, test, and upgrade JARVIS itself. At startup, JARVIS warms the persistent workspace index asynchronously and starts an incremental filesystem watcher, keeping repository-aware context responsive while allowing MCP/Web to become ready immediately.
See .obsidian-wiki/JARVIS-MCP-Architecture.md for the living architecture and .obsidian-wiki/00_System/log.md for the development timeline.
Session resilience /
JARVIS/jarvis resilience paths keep runtime ownership explicit across MCP listener restarts, STDIO replacement, and Secure MCP Tunnel lifecycle changes. The Desktop and PowerShell tunnel publishers share the same ownership lock and critical section; stale ownership is reclaimed only after process identity verification, and unverifiable liveness fails closed. Acceptance coverage also verifies that parallel workspace sessions retain independent session/workspace ownership and that the production launcher does not start a second tunnel client when another owner is active.
For operators, the tunnel client is resolved from JARVIS_TUNNEL_CLIENT_PATH when configured; otherwise use the packaged/default tunnel-client location. The diagnostic command is:
$tc = if ($env:JARVIS_TUNNEL_CLIENT_PATH) { $env:JARVIS_TUNNEL_CLIENT_PATH } else { 'tunnel-client.exe' }
& $tc doctor --profile jarvis --profile-dir $profile --explainThe MCP listener uses an ephemeral loopback port where the runtime owns the listener directly; fixed nonzero listener ports are not required by the resilience workflow.
Complete MCP tool catalog (239 total definitions; 194 advertised by default; 200 with Codex enabled)
This complete index is generated from ToolRegistry.listAll(), not copied from an older release document. The default tools/list surface advertises only operational or dependency-gated definitions; planned and feature-disabled definitions remain visible here without being advertised. Enabling Codex delegation adds its six operational definitions to the advertised surface.
# | Tool | Permission | Advertised | Delivery | Runtime evidence | Runtime description |
1 |
| READ | default | operational | service_dispatch | List registered project workspaces available to jarvis. Legacy explicitly registered drive roots may also appear as kind=machine_root. |
2 |
| WRITE | default | operational | service_dispatch | Register an existing project directory by absolute path. parentWorkspaceId is optional and retained only for legacy machine-root-relative registration. Idempotent for the same path. |
3 |
| READ | default | operational | service_dispatch | Return the configured workspace summary. |
4 |
| READ | default | operational | service_dispatch | List a bounded workspace tree. Absolute path does not require workspaceId. |
5 |
| READ | default | operational | service_dispatch | Return a bounded project snapshot without source contents. |
6 |
| READ | default | operational | service_dispatch | Read a workspace file as UTF-8 text or as an image/binary payload. Absolute paths (C:...) do not require workspaceId. For large files or an unknown location, prefer search_text first and then read_file_page for the relevant range instead of reading the whole file. |
7 |
| READ | default | operational | service_dispatch | Read up to twenty bounded workspace files in parallel. Absolute paths do not require workspaceId. For large files, locate text with search_text and page with read_file_page instead of loading entire files. |
8 |
| READ | default | operational | service_dispatch | Search workspace filenames with automatic context-economy filters; set includeIgnored for an explicit full path search. Absolute path does not require workspaceId. |
9 |
| READ | default | operational | service_dispatch | Preferred tool to locate relevant code/lines before reading files. Searches workspace text using direct ripgrep arguments with automatic binary/generated filters; set includeIgnored for an explicit full path search. Absolute path does not require workspaceId. Follow with read_file_page for large files. |
10 |
| READ | default | operational | service_dispatch | Inspect parsed read-only Git status. For writes (init, add, commit, remote, push, rm, clean, reset) use the git tool. |
11 |
| READ | default | operational | service_dispatch | Return a bounded read-only Git diff. For writes use the git tool. |
12 |
| READ | default | operational | service_dispatch | Return bounded structured Git history. For writes use the git tool. |
13 |
| EXECUTE | default | operational | service_dispatch | Run a Git subcommand with a separate args array. With Full Bypass OFF, Full Access runs ordinary read and non-destructive Git mutations without confirmation while destructive/data-loss forms, scope overrides, aliases, unsafe pathspecs, unknown commands, and destructive remote/history rewrites remain guarded or denied. Trusted Full Bypass skips jarvis approval, command-policy, and Active Project scope checks, including explicitly absolute outside paths, without bypassing Git or OS errors. Do not wrap Git in PowerShell/cmd. |
14 |
| WRITE | default | operational | service_dispatch | Create or replace a UTF-8 text file and missing parents. Balanced/Safe refuse existing targets unless overwriteExisting is explicit; Full may replace an existing target without a confirmation prompt and still creates a checkpoint. Prefer edit_file for narrow repairs. Use this instead of shell scripts that call fs.writeFile, writeFileSync, Set-Content, or equivalent when the task is simply to create or replace guarded text. |
15 |
| WRITE | default | operational | service_dispatch | Apply reviewed whole-file replacement content to at most twenty files. Existing targets are checkpointed first; Full profile does not prompt for non-destructive replacement. Prefer edit_file for narrow repairs. Use this instead of shell-generated whole-file rewrites when several reviewed text files must change. |
16 |
| WRITE | default | operational | service_dispatch | First choice for narrow source, config, and text repairs. Replaces exact text only when the expected occurrence count matches, checkpoints the original, and refuses conflicts instead of rewriting an unverified whole file. Use edit_file instead of shell, node -e, python -c, PowerShell Set-Content, or inline filesystem scripts when a guarded text edit can express the change. Full Access performs ordinary edits without a confirmation prompt; destructive deletion remains separately guarded. |
17 |
| WRITE | default | operational | service_dispatch | Move a file or directory, creating missing destination parents. With Full Bypass OFF, Full Access performs ordinary in-project moves without a confirmation prompt while conflicting or destructive forms remain policy-gated. Trusted Full Bypass skips jarvis approval/scope checks for explicit absolute outside paths; OS/filesystem errors still apply. |
18 |
| WRITE | default | operational | service_dispatch | Copy a file or directory within one workspace, creating missing destination parents. |
19 |
| DANGEROUS | default | operational | service_dispatch | Delete one file or empty directory. With Full Bypass OFF, eligible in-project targets move to Recovery Trash and exact safe targets can use scoped auto-approval; critical paths, roots, non-empty directories, ambiguous paths, and mismatched workspaces remain guarded. Trusted Full Bypass skips jarvis approval/scope checks and permits an exact absolute outside target, which is deleted without Recovery Trash; root and non-empty-directory input guards still apply. |
20 |
| READ | default | operational | service_dispatch | List trusted Recovery Trash entries for one workspace, including deleted items, binary pre-replacement backups, original paths, timestamps, payload availability, and the local Recovery Trash root. |
21 |
| WRITE | default | operational | service_dispatch | Restore one Recovery Trash item to its original path. Deleted-item restores refuse existing targets. A pre-replacement restore first backs up the current live version for undo, then restores the older binary or text payload. Full runs recoverable restores without an extra prompt; stricter profiles may require confirmation. The operation remains scoped to the recorded workspace. |
22 |
| READ | default | operational | service_dispatch | List encrypted pre-mutation checkpoints for one workspace without returning saved file content. |
23 |
| WRITE | default | operational | service_dispatch | Restore a reviewed pre-mutation checkpoint. Standard mode requires explicit confirmation; trusted Full Bypass skips the jarvis confirmation gate. A new rollback checkpoint is created before replacing current content when the target is inside a recoverable workspace. |
24 |
| EXECUTE | default | operational | service_dispatch | Immediate-return managed process launcher for real executables and long-lived processes. With Full Bypass OFF, inline text-file rewrites must use edit_file/apply_patch/write_file and risky commands, scope changes, or permission-profile ASK decisions require confirmation. Trusted Full Bypass skips jarvis command/profile/scope approval, including an explicitly absolute cwd outside the Active Project; input validation, executable availability, OS rights, and exact process ownership still apply. Starts one executable with separate arguments and returns processId as soon as the child is spawned; it never waits for command completion. Follow with process_status/process_logs/process_stop. For restart-safe durable work, use shell, whose MCP run mode is forced to background. |
25 |
| READ | default | operational | service_dispatch | List managed process handles owned by this client in a workspace, including launches whose response was cancelled. |
26 |
| READ | default | operational | service_dispatch | Read one status snapshot for an owned process handle. Do not tight-poll this tool; use project_* for normal project verification, or shell background + durable task_id for work expected to exceed ~5 minutes. |
27 |
| READ | default | operational | service_dispatch | Read bounded logs for an owned process handle. Prefer one bounded log read after meaningful progress rather than repeated status polling. |
28 |
| EXECUTE | default | operational | service_dispatch | Stop an owned managed process tree after explicit chat confirmation in standard mode. Trusted Full Bypass skips the jarvis confirmation gate; exact process ownership still applies. |
29 |
| EXECUTE | default | operational | service_dispatch | Immediate-return launcher for the detected project dev command. In standard mode the gateway previews the exact executable/argv for host approval and re-resolves it immediately before spawn; any change requires fresh approval. Trusted Full Bypass skips the jarvis approval boundary. Project-owned script bodies remain opaque and are not covered by Recovery Trash. |
30 |
| EXECUTE | default | operational | service_dispatch | Immediate-return launcher for the detected project test command. In standard mode the gateway previews the exact executable/argv for host approval and re-resolves it immediately before spawn; any change requires fresh approval. Trusted Full Bypass skips the jarvis approval boundary. Project-owned script bodies remain opaque and are not covered by Recovery Trash. |
31 |
| EXECUTE | default | operational | service_dispatch | Immediate-return launcher for the detected project lint command. In standard mode the gateway previews the exact executable/argv for host approval and re-resolves it immediately before spawn; any change requires fresh approval. Trusted Full Bypass skips the jarvis approval boundary. Project-owned script bodies remain opaque and are not covered by Recovery Trash. |
32 |
| EXECUTE | default | operational | service_dispatch | Immediate-return launcher for the detected project typecheck command. In standard mode the gateway previews the exact executable/argv for host approval and re-resolves it immediately before spawn; any change requires fresh approval. Trusted Full Bypass skips the jarvis approval boundary. Project-owned script bodies remain opaque and are not covered by Recovery Trash. |
33 |
| EXECUTE | default | operational | service_dispatch | Immediate-return launcher for the detected project build command. In standard mode the gateway previews the exact executable/argv for host approval and re-resolves it immediately before spawn; any change requires fresh approval. Trusted Full Bypass skips the jarvis approval boundary. Project-owned script bodies remain opaque and are not covered by Recovery Trash. |
34 |
| READ | Codex opt-in | operational | service_dispatch | Report local Codex installation and capabilities without credential inspection. |
35 |
| EXECUTE | Codex opt-in | operational | service_dispatch | Delegate an instruction to the local Codex CLI in the Active Project. Starting Codex requires explicit chat confirmation and host approval in standard mode; trusted Full Bypass skips those jarvis application checks without forging userConfirmed. |
36 |
| READ | Codex opt-in | operational | service_dispatch | List local Codex task handles owned by this client, including launches whose response was cancelled. |
37 |
| READ | Codex opt-in | operational | service_dispatch | Read status for an owned Codex task. |
38 |
| READ | Codex opt-in | operational | service_dispatch | Read bounded logs for an owned Codex task. |
39 |
| EXECUTE | Codex opt-in | operational | service_dispatch | Stop an owned Codex task process after explicit chat confirmation in standard mode. Trusted Full Bypass skips the jarvis confirmation gate; task ownership still applies. |
40 |
| EXECUTE | default | operational | service_dispatch | Non-blocking command runner for real command execution, builds/tests, package managers, and system operations. Never use shell as a source/config/text editor. For any direct text-file change, call edit_file first; use apply_patch for reviewed whole-file or multi-file replacements and write_file for file creation/replacement. Inline Node/Python/PowerShell/sed commands that rewrite text files are rejected before native approval so the client can route to the guarded file tools instead. MCP run calls are ALWAYS forced to execution=background, even if a client requests foreground or auto, so the call returns a task_id immediately instead of waiting for command completion. Follow with status/logs/result; wait uses the user-configurable MCP poll window (5-60 seconds, default 5). After one or two checks still show running, do not keep polling in the same chat turn: preserve task_id and return control so the durable task can continue without risking a ChatGPT turn timeout. When the host turn must yield while a durable task is still running, checkpoint its task_id and use the active scheduled-continuation handoff instead of abandoning the goal. With Full Bypass OFF, Full Access runs ordinary policy-allowed commands without confirmation while destructive, broad, recursive, critical, outside-project, or unparseable forms retain normal approval/command policy. Trusted Full Bypass skips jarvis approval, command-policy, Active Project, goalLease, and allowed-root checks, including an explicitly absolute cwd outside the project; input validation, executable availability, Windows ACL/UAC, and child-process failures still apply. dry_run and task observation are non-mutating. |
41 |
| READ | default | operational | service_dispatch | Default for web-page DOM work inside managed Chrome. Call list_tabs first, select the exact returned tab_id by URL/title, and pass that tab_id to every query, click, type, navigate, evaluate, wait, screenshot, close, or steps call. If no safe matching tab exists, call new_tab and use its returned ID. Target order and the OS-active tab are never ownership signals. Never navigate through the browser address bar with computer_use/accessibility/input_event. Protected ChatGPT tab mutations additionally require allow_protected_tab_action=true plus explicit user confirmation. |
42 |
| EXECUTE | default | operational | service_dispatch | Codex-style native Windows computer use for testing desktop apps. Take annotated screenshots, inspect semantic controls, and operate by semantic target, numbered visual mark, or explicit coordinates. Routes through Accessibility first and uses guarded pointer/keyboard input only when needed. Supports click, typing, keys, hotkeys, scroll, drag, pointer movement, and window activation. For web navigation, do not focus/type into a browser address bar; use dom_cdp list_tabs/new_tab plus an explicit tab_id. |
43 |
| READ | default | operational | service_dispatch | Semantic native Windows UI tool. Inspect UI trees and named controls, then click, focus, read or set values, select controls and menus, or manage a native element. Prefer shell for direct system work and dom_cdp for web pages. |
44 |
| EXECUTE | default | operational | service_dispatch | Low-level keyboard and pointer fallback. Use only when DOM/CDP and Accessibility cannot operate the target. Supports text, keys, mouse movement, clicks, drag, scroll, held buttons, release_all, and batched sequences. For web navigation, do not focus/type into a browser address bar; use dom_cdp list_tabs/new_tab plus an explicit tab_id. |
45 |
| READ | default | operational | service_dispatch | Visual and OCR fallback for content unavailable through DOM or Accessibility. Capture a display, window, or region, or run local Vision OCR. It never clicks or types. |
46 |
| READ | default | operational | service_dispatch | Capture a local Windows screen/region/window and return a short-lived Set-of-Marks observation with numbered bounds, a content hash, and an annotated PNG. This tool only observes; use ui_target_action for a separately gated action. |
47 |
| EXECUTE | default | operational | service_dispatch | Act on one mark from a current vision_annotated_capture observation. The observation ID, optional hash, TTL, workspace owner, and current Accessibility element are checked before the action is sent. |
48 |
| EXECUTE | default | operational | service_dispatch | Direct native Windows window management. List, inspect, activate, move, resize, minimize, maximize, restore, or close windows without raw coordinates when a window operation is sufficient. |
49 |
| READ | default | operational | service_dispatch | Diagnostics only. Check all jarvis backends or one public tool after a failure, when asked for status, or while diagnosing permissions. Do not use as a preflight before normal work. |
50 |
| READ | default | operational | service_dispatch | Read-only system information: OS, CPU, memory, disks, battery, uptime, and top processes by memory. Use for environment checks and diagnostics. |
51 |
| EXECUTE | default | operational | service_dispatch | Show a Windows notification (toast when BurntToast is installed, balloon otherwise). Use to tell the user when a long task finishes. |
52 |
| EXECUTE | default | operational | service_dispatch | Open a native Windows file open/save dialog and return the chosen path(s). The dialog does not read or write files itself; use the guarded file tools afterwards. |
53 |
| EXECUTE | default | operational | service_dispatch | Read or write the Windows clipboard (text, or PNG image as base64). Use get_text/get_image to read and set_text to write. |
54 |
| READ | default | operational | service_dispatch | Fetch an http/https URL (GET/POST/PUT/DELETE/HEAD) with bounded size and timeout. In standard mode every POST, PUT, or DELETE requires explicit chat confirmation and host approval; trusted Full Bypass skips jarvis approval. dry_run remains safe. Returns status, headers, and text or base64 body. |
55 |
| EXECUTE | default | operational | service_dispatch | Record the microphone to a WAV file or play a local audio file through MCI. In standard mode recording requires the host-selected Active Project workspaceId and explicit confirmation; trusted Full Bypass skips jarvis approval/scope checks. Existing in-workspace outputs use Recovery Trash before replacement when available. record is synchronous and limited to 600 seconds. Use stop to abort an ongoing record/play. |
56 |
| EXECUTE | default | operational | service_dispatch | Record the screen to an MP4 using ffmpeg gdigrab (requires ffmpeg on PATH). In standard mode starting a recording requires the host-selected Active Project workspaceId and explicit confirmation; trusted Full Bypass skips jarvis approval/scope checks. Existing in-workspace outputs use Recovery Trash before replacement when available. start spawns a background capture, status checks it, stop finalizes the file. Recording stops automatically after 3600 seconds. |
57 |
| WRITE | default | operational | service_dispatch | Automate Excel, Word, PowerPoint, or Outlook through COM. In standard mode every write, replace, merge, or save_as action requires an Active Project workspaceId, explicit chat confirmation, and host approval. Trusted Full Bypass skips jarvis approval/scope checks without forging userConfirmed. Existing in-workspace targets use Recovery Trash before replacement when available. Requires Microsoft Office installed. |
58 |
| EXECUTE | default | operational | service_dispatch | Manage Windows scheduled tasks with schtasks.exe. list is read-only; in standard mode create, run, and delete require explicit chat confirmation and host approval. Trusted Full Bypass skips jarvis approval without forging userConfirmed. |
59 |
| EXECUTE | default | operational | service_dispatch | Non-blocking WSL2 developer runner for one Linux executable plus argv; shell command strings are not accepted. Do not use wsl_exec as a source/config/text editor. For any direct text-file change, call edit_file first; use apply_patch for reviewed whole-file or multi-file replacements and write_file for file creation/replacement. Inline Node/Python/PowerShell-style rewrites and sed in-place edits are rejected before native approval so the client can route to guarded file tools. MCP run calls are ALWAYS forced to execution=background, even if a client requests foreground or auto, and return a task_id immediately. Follow with status/logs/result; wait uses the user-configurable MCP poll window (5-60 seconds, default 5). After one or two checks still show running, do not keep polling in the same chat turn: preserve task_id and return control so the durable task can continue without risking a ChatGPT turn timeout. When the host turn must yield while a durable task is still running, checkpoint its task_id and use the active scheduled-continuation handoff instead of abandoning the goal. With Full Bypass OFF, Full Access runs ordinary WSL commands without confirmation while destructive, broad, recursive, outside-project, or unparseable forms retain normal approval/command policy. Trusted Full Bypass skips jarvis approval, command-policy, Active Project, goalLease, and allowed-root checks, including an explicitly requested external cwd; WSL availability, argv validation, Linux permissions, and process failures still apply. |
60 |
| READ | default | operational | service_dispatch | Translate paths and inspect metadata between a registered Windows workspace and WSL without exposing raw \wsl$ read/write access. |
61 |
| READ | default | operational | service_dispatch | List the union of bundled skills and every discovered machine-global or active-workspace skill from Cursor, Claude, Agents, Codex, the Codex plugin cache, GitHub workspace roots, and jarvis settings. Nested and symlinked skill collections are included. Filter with query or source. |
62 |
| READ | default | operational | service_dispatch | Read a local skill SKILL.md (or a relative file inside the skill folder). Prefer the source-qualified id returned by skills_list; an unambiguous bare name or $name is also accepted. Follow the skill instructions with jarvis tools and mcp_call. |
63 |
| READ | default | operational | service_dispatch | List local MCP servers discovered from Cursor, Claude Desktop, and jarvis settings. This inspection is read-only and does not flatten child tools into the jarvis catalog. |
64 |
| READ | default | operational | service_dispatch | Connect to one local MCP server (if needed) and return its tool names, descriptions, and input schemas. This operation only inspects the child tool catalog. |
65 |
| DANGEROUS | default | operational | service_dispatch | Call a tool on a discovered local MCP server. Child side effects and filesystem/network scope are controlled by that child server, so standard mode treats every mcp_call as opaque mutation and requires explicit chat plus host exact-action approval. Trusted Full Bypass skips jarvis application approval; the child server still enforces its own policy. |
66 |
| READ | default | operational | service_dispatch | Aggregate ranked workspace context with snippets, symbols, Git/test relevance, economy metadata, and continuation; automatic discovery can be explicitly expanded. |
67 |
| READ | default | operational | service_dispatch | Continue a workspace_context result without discarding unreturned candidates. |
68 |
| READ | default | operational | service_dispatch | Enumerate workspace files with full access by default; set includeIgnored false to use the persistent automatic index. |
69 |
| READ | default | operational | deterministic_operation | Continue a workspace_full_scan result page. |
70 |
| READ | default | operational | service_dispatch | Return workspace identity and project snapshot metadata without source contents. |
71 |
| READ | default | operational | service_dispatch | Search text and filenames across one or all registered workspaces with automatic economy filters or an explicit includeIgnored override. |
72 |
| READ | default | operational | service_dispatch | Read many workspace files in parallel while preserving one result or error per requested path. |
73 |
| READ | default | operational | service_dispatch | Preferred reader for large files after search_text identifies the relevant area. Reads a deterministic line chunk with explicit continuation instead of silently truncating or loading the whole file. |
74 |
| READ | default | operational | service_dispatch | Continue read_file_page from the next deterministic line chunk only when more surrounding context is needed; avoid re-reading earlier pages. |
75 |
| READ | default | operational | service_dispatch | Build or refresh the persistent workspace index using automatic context filters unless ignored paths are explicitly included. |
76 |
| READ | default | operational | service_dispatch | Return persistent index metadata and lossless watcher queue telemetry. |
77 |
| READ | default | operational | service_dispatch | Watch all workspace paths and incrementally re-index only changed paths with configurable debounce/concurrency. |
78 |
| READ | default | operational | service_dispatch | Stop a workspace watcher after draining all queued path updates. |
79 |
| READ | default | operational | service_dispatch | Create a concise same-chat recovery message from the real phase tracker, current git status/diff, and durable background task IDs. Use only when the user requests a handoff or an unavoidable client/platform interruption requires recovery; never trigger it merely because elapsed time passed. If a tool schema looks stale, Refresh connector first; open a new chat only if refresh does not fix it. |
80 |
| EXECUTE | default | operational | service_dispatch | Run the detected project typecheck only when the current git status/diff fingerprint changed. Starting a new verification process requires explicit user confirmation in standard mode; trusted Full Bypass skips that jarvis gate. Returns cache=hit when unchanged and cache=miss after a new verification. Prefer this during iterative edits; use project_test/project_lint/project_build only when that specific verification is needed. For full suites or packaging expected to exceed ~5 minutes, launch a durable shell background task and record its task_id in the tracker. |
81 |
| WRITE | default | operational | service_dispatch | Immediate-return durable goal create/resume and lease acquisition. Unfinished goals default to scheduledContinuation=auto: the client must automatically load/follow the bundled jarvis-scheduled-continuation skill, keep exactly one native one-time cloud successor after a real checkpoint, continue useful work without waiting for the user to type continue/ทำต่อ, and stop scheduling only when the goal is terminal or scheduling is explicitly disabled. Native ChatGPT task creation remains host-owned; this tool never claims that a task was created. |
82 |
| READ | default | operational | service_dispatch | Read the latest durable goal snapshot without changing state or returning a lease token. |
83 |
| WRITE | default | operational | service_dispatch | Atomically checkpoint durable goal progress using the current lease and expected revision. For an active goal using the default automatic continuation contract, a successful real checkpoint is the handoff point where the client must ensure exactly one native one-time cloud successor through jarvis-scheduled-continuation before yielding; never wait for the user to type continue/ทำต่อ. |
84 |
| WRITE | default | operational | service_dispatch | Finish the local durable goal using lease/revision compare-and-swap. It must be called before any completion report, even when scheduling was disabled or the user requested no more successors. If scheduledTaskCancellation requests delete_native_task, delete that exact task through the native ChatGPT Scheduled Task host, record its native deletion receipt, and verify status=cancelled before reporting cancellation success. |
85 |
| WRITE | default | operational | service_dispatch | Cancel a durable goal independently of any scheduled successor. It records the goal as cancelled, aborts in-flight fenced MCP requests for that goal, and attempts to stop every tracked process, Codex task, and shell task across the current runtime and durable task store; inspect requestCancellation, taskCancellations, and allRequestsStopped/allTasksStopped for unresolved work. If scheduledTaskCancellation requests delete_native_task, use cancel_scheduled_continuation separately and complete the exact native ChatGPT host deletion receipt. |
86 |
| READ | default | operational | service_dispatch | List a bounded set of durable goals owned by the current stable MCP client, optionally filtered by workspace/status. |
87 |
| WRITE | default | operational | service_dispatch | Checkpoint and reserve exactly one current-chat cloud successor with an adaptive delay between 2 and 25 minutes. Omitted delay defaults to the fail-safe +2-minute handoff; a healthy current run may explicitly choose a longer 5/10/25-minute watchdog. This workflow never creates or deletes the native task itself. |
88 |
| WRITE | default | operational | service_dispatch | Record host-owned cloud one-time task create, same-task reschedule, consumed-run reconciliation, or cancellation receipts. A consumed receipt requires exact native host run evidence and means only that the one-time task is no longer pending; it does not mean the goal work completed. Cancelled is accepted only with a matching native ChatGPT host deletion receipt; a model assertion is not cancellation proof. The stored native task ID is immutable across reschedules. |
89 |
| WRITE | default | operational | service_dispatch | Scheduled-wake entrypoint. Claim before workspace mutation; a confirmed cloud wake up to 120 seconds early is accepted so native host jitter does not consume the one-time task without handoff. If native task creation was never confirmed, returns receipt_required for reconciliation. A one-time task that is firing is treated as a consumed wake ticket: on an active-worker collision, claim atomically supersedes that ticket and returns successor_required with a fresh +2-minute cloud scheduleRequest. Create that fresh successor and let the current wake finish naturally; never re-arm the firing task. If the outcome is terminal_noop, let the already-firing host task return naturally; do not delete, disable, pause, or reschedule it. Do not mutate the workspace or mark the goal terminal on collision. |
90 |
| READ | default | operational | service_dispatch | Read one scheduled-continuation snapshot by continuation ID or the latest record for a goal. A healthy current run keeps its adaptive watchdog unless a real turn-yield signal requires same-task +2 handoff. |
91 |
| WRITE | default | operational | service_dispatch | For an enumerated handoff-risk signal, including a turn that is about to end while the goal is unfinished, move the exact existing cloud one-time native task to now+2 minutes. No replacement task is created. |
92 |
| WRITE | default | operational | service_dispatch | Cancel one still-pending scheduled successor independently of its goal. Identify it by continuationId or the latest record for a goal, then use the returned cancellation instruction to delete the exact pending native ChatGPT Scheduled Task and record its host receipt. Never treat pausing/disabling an already-fired current wake as deletion or completion proof. This does not cancel the durable goal or stop its running tasks. |
93 |
| READ | default | operational | service_dispatch | Search indexed symbols across the workspace. |
94 |
| READ | default | operational | service_dispatch | Find deterministic symbol definitions. |
95 |
| READ | default | operational | service_dispatch | Find textual and indexed references to a symbol. |
96 |
| READ | default | operational | service_dispatch | Find interface and class implementations. |
97 |
| READ | default | operational | service_dispatch | Return a deterministic call hierarchy approximation. |
98 |
| READ | default | operational | service_dispatch | Return indexed imports and exports for a module. |
99 |
| READ | default | operational | service_dispatch | Return package and module dependency metadata. |
100 |
| READ | default | operational | service_dispatch | Return the workspace module graph. |
101 |
| READ | default | operational | service_dispatch | Search indexed TypeScript, JavaScript, and Python types. |
102 |
| READ | default | operational | service_dispatch | Combine definition, references, imports, tests, and recent context. |
103 |
| READ | default | operational | deterministic_operation | Explain ranking signals without removing lower-ranked context. |
104 |
| READ | default | operational | service_dispatch | Gather deterministic debugging context and continuation metadata. |
105 |
| READ | default | operational | service_dispatch | Gather code-review context. |
106 |
| READ | default | operational | service_dispatch | Gather changed files, symbols, dependencies, and tests. |
107 |
| READ | default | operational | service_dispatch | Gather context around a symbol. |
108 |
| READ | default | operational | service_dispatch | Gather relevant test context. |
109 |
| READ | default | operational | service_dispatch | Gather dependency-related context. |
110 |
| READ | default | operational | service_dispatch | Gather Git status, diff, and history context. |
111 |
| READ | default | operational | service_dispatch | Gather frontend project context. |
112 |
| READ | default | operational | service_dispatch | Gather backend project context. |
113 |
| READ | default | operational | deterministic_operation | Deprecated alias of jarvis_tool_route: classify a prompt with a deterministic, overridable route. |
114 |
| READ | default | operational | deterministic_operation | List built-in and user recipe names. |
115 |
| READ | default | operational | deterministic_operation | Describe a recipe plan and permissions. |
116 |
| EXECUTE | default | operational | deterministic_operation | Preview or run a deterministic recipe plan. |
117 |
| READ | default | operational | deterministic_operation | Return a no-side-effect execution preview. |
118 |
| READ | no | planned | service_dispatch | Review current Git changes and affected context. |
119 |
| READ | default | operational | service_dispatch | Find symbols in changed files. |
120 |
| READ | no | planned | service_dispatch | Find modules affected by current changes. |
121 |
| READ | no | planned | service_dispatch | Return relevant recent Git history. |
122 |
| READ | no | planned | service_dispatch | Return line ownership context for a file. |
123 |
| READ | no | planned | service_dispatch | Discover project tests without imposing an execution limit. |
124 |
| EXECUTE | no | planned | service_dispatch | Plan or run tests affected by changed files. |
125 |
| READ | no | planned | service_dispatch | Summarize recorded test failures. |
126 |
| READ | no | planned | service_dispatch | Return coverage context when project tooling provides it. |
127 |
| READ | no | planned | service_dispatch | Return recent test execution history. |
128 |
| READ | default | operational | deterministic_operation | Return shared cache hit/miss telemetry. |
129 |
| WRITE | default | operational | deterministic_operation | Clear safe local runtime caches. |
130 |
| WRITE | default | operational | deterministic_operation | Invalidate cache entries for a path or workspace. |
131 |
| READ | default | operational | deterministic_operation | List registered lifecycle hooks. |
132 |
| WRITE | default | operational | deterministic_operation | Register a deterministic lifecycle hook descriptor. |
133 |
| WRITE | default | operational | deterministic_operation | Remove a lifecycle hook descriptor. |
134 |
| READ | default | operational | service_dispatch | Match relevant local skills without loading all skill text. |
135 |
| READ | default | operational | service_dispatch | Load a selected local skill by identifier. |
136 |
| WRITE | no | feature_disabled | truthful_unavailable | Register a declared plugin descriptor after validation and permission evaluation. |
137 |
| READ | no | feature_disabled | truthful_unavailable | List installed and enabled plugins. |
138 |
| WRITE | no | feature_disabled | truthful_unavailable | Enable an installed plugin. |
139 |
| WRITE | no | feature_disabled | truthful_unavailable | Disable an installed plugin. |
140 |
| DANGEROUS | no | feature_disabled | truthful_unavailable | Remove an installed plugin. |
141 |
| READ | default | operational | deterministic_operation | Return persisted development-session context. |
142 |
| WRITE | default | operational | deterministic_operation | Persist a development-session checkpoint. |
143 |
| READ | default | operational | deterministic_operation | Resume a persisted session context. |
144 |
| READ | default | operational | deterministic_operation | Return session checkpoints and decisions. |
145 |
| READ | default | operational | deterministic_operation | Select compact, normal, verbose, or stream formatting. |
146 |
| READ | default | operational | service_dispatch | Combine DOM, console, network, URL, and screenshot metadata. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
147 |
| READ | default | operational | service_dispatch | Gather deterministic UI debugging context. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
148 |
| READ | default | operational | service_dispatch | Capture a structured UI state. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
149 |
| READ | default | operational | service_dispatch | Inspect form controls and values metadata. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
150 |
| READ | default | dependency_gated | truthful_unavailable | Summarize browser network context when a retained CDP network event stream is available. |
151 |
| READ | default | dependency_gated | truthful_unavailable | Summarize browser console context when a retained CDP Runtime/Log event stream is available. |
152 |
| READ | default | operational | service_dispatch | Combine browser diagnostics for one request. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
153 |
| READ | default | operational | truthful_unavailable | Inspect Windows environment metadata. |
154 |
| READ | default | operational | truthful_unavailable | Inspect Windows service metadata. |
155 |
| READ | default | operational | truthful_unavailable | Inspect process-tree context. |
156 |
| READ | default | operational | truthful_unavailable | Inspect local listening-port context. |
157 |
| READ | default | operational | truthful_unavailable | Inspect registry context through the Windows capability boundary. |
158 |
| READ | default | operational | truthful_unavailable | Inspect Windows event-log context. |
159 |
| READ | default | operational | truthful_unavailable | Inspect installed runtimes and package managers. |
160 |
| READ | default | operational | truthful_unavailable | Resolve executable and PATH context. |
161 |
| READ | default | operational | truthful_unavailable | Inspect startup configuration context. |
162 |
| READ | default | operational | deterministic_operation | Discover external MCP servers without flattening native tools. |
163 |
| READ | default | operational | deterministic_operation | Return external MCP connection health. |
164 |
| READ | default | dependency_gated | deterministic_operation | List resources exposed by connected MCP servers when the child server supports resources/list. |
165 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Create a visible managed runtime task. |
166 |
| READ | no | feature_disabled | truthful_unavailable | Read managed task state. |
167 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Cancel a managed runtime task. |
168 |
| READ | no | feature_disabled | truthful_unavailable | Read a managed task result. |
169 |
| READ | no | feature_disabled | truthful_unavailable | List managed runtime tasks. |
170 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Delegate a task through a policy/audit adapter. |
171 |
| READ | no | feature_disabled | truthful_unavailable | Read delegated agent state. |
172 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Cancel a delegated agent task. |
173 |
| READ | no | feature_disabled | truthful_unavailable | Read delegated agent result. |
174 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Run isolated read-only agent tasks with collision metadata. |
175 |
| READ | default | operational | deterministic_operation | Evaluate an action class without limiting allowed context reads. |
176 |
| READ | default | operational | deterministic_operation | Return the active Permission v2 profile. |
177 |
| READ | no | feature_disabled | deterministic_operation | Query structured activity/log metadata with correlation IDs. |
178 |
| READ | no | feature_disabled | deterministic_operation | Return Live Logs pipeline health and source status. |
179 |
| READ | no | feature_disabled | deterministic_operation | Return runtime performance telemetry. |
180 |
| READ | default | operational | deterministic_operation | Return context discovery, deduplication, ledger, and token-efficiency telemetry. |
181 |
| READ | default | operational | deterministic_operation | Return the cheapest deterministic execution plan and reason. |
182 |
| READ | default | operational | service_dispatch | Return a traversable repository structural map. |
183 |
| READ | default | operational | service_dispatch | Return optional import, caller, type, test, and change references. |
184 |
| READ | default | operational | deterministic_operation | Return reconnect, retry, continuation, cache, and worker recovery state. |
185 |
| READ | default | operational | deterministic_operation | List versioned tool schema metadata. |
186 |
| WRITE | no | feature_disabled | deterministic_operation | Register a backward-compatible tool schema descriptor. |
187 |
| READ | default | operational | deterministic_operation | Discover capability categories without requiring every full schema. |
188 |
| READ | default | operational | deterministic_operation | Search tools, tags, phases, and descriptions deterministically. |
189 |
| READ | default | operational | deterministic_operation | Return a bounded ranked tool set using deterministic scoring with optional local rerank fallback. |
190 |
| READ | default | operational | deterministic_operation | Describe one tool contract on demand. |
191 |
| READ | default | operational | deterministic_operation | List tool categories and counts. |
192 |
| READ | default | operational | deterministic_operation | Find the best local tool/function candidates for a prompt. |
193 |
| READ | default | operational | deterministic_operation | List stable shorthand aliases and their primitive tool targets. |
194 |
| READ | default | dependency_gated | deterministic_operation | Describe the additive MCP hub boundary without flattening child tools or retaining credentials. |
195 |
| READ | default | operational | deterministic_operation | Run the unified deterministic development-context facade. |
196 |
| READ | default | operational | deterministic_operation | Return inspectable developer automation recipes. |
197 |
| READ | default | operational | service_dispatch | Capture screenshot metadata for visual validation. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
198 |
| READ | default | operational | deterministic_operation | Compare screenshot metadata or supplied artifacts. |
199 |
| READ | default | operational | service_dispatch | Return a structured DOM snapshot. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
200 |
| READ | default | operational | service_dispatch | Return layout metadata for visual validation. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
201 |
| READ | default | operational | service_dispatch | Combine screenshot, DOM, layout, console, and network references. Requires an exact dom_cdp tab_id from list_tabs or new_tab; never uses the active/first tab. |
202 |
| READ | default | operational | service_dispatch | Inspect workbook sheets, used ranges, and a bounded sample through Excel COM. |
203 |
| READ | no | feature_disabled | deterministic_operation | Compare workbook layout metadata through an optional spreadsheet plugin. |
204 |
| READ | no | feature_disabled | deterministic_operation | Render an Excel preview through an optional spreadsheet plugin. |
205 |
| READ | default | dependency_gated | truthful_unavailable | Inspect PDF page structure and text through the local PDF provider. |
206 |
| READ | no | feature_disabled | deterministic_operation | Compare PDF page metadata through an optional PDF plugin. |
207 |
| READ | no | feature_disabled | truthful_unavailable | Read project intelligence conventions. |
208 |
| WRITE | no | feature_disabled | truthful_unavailable | Update project intelligence conventions. |
209 |
| READ | default | operational | deterministic_operation | Build a structured cross-agent handoff bundle from real workspace, Git, and context services. |
210 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Run or preview a benchmark scenario. |
211 |
| READ | no | feature_disabled | truthful_unavailable | Return benchmark and regression results. |
212 |
| EXECUTE | default | dependency_gated | truthful_unavailable | Run an artifact-based Windows Sandbox job with networking disabled and read-only mapped input. |
213 |
| EXECUTE | default | dependency_gated | deterministic_operation | Watch an allowlisted user-mode ETW or Windows Event Log diagnostic stream. |
214 |
| READ | default | dependency_gated | deterministic_operation | Return bounded crash and service-diagnostic context from allowlisted user-mode sources. |
215 |
| READ | default | dependency_gated | truthful_unavailable | Read diagnostics from an owned language-server child process. |
216 |
| WRITE | default | dependency_gated | truthful_unavailable | Create a cross-file LSP rename edit plan before any workspace write. |
217 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Attach a DAP client only to an owned workspace debug adapter. |
218 |
| EXECUTE | no | feature_disabled | truthful_unavailable | Perform a bounded DAP stepping/read operation in an owned debug session. |
219 |
| WRITE | default | dependency_gated | deterministic_operation | Create a confined, ledger-owned Git worktree for isolated agent work with collision metadata. |
220 |
| DANGEROUS | default | dependency_gated | deterministic_operation | Remove a ledger-owned Git worktree after dry-run and standard-mode confirmation; trusted Full Bypass skips jarvis approval. |
221 |
| READ | default | dependency_gated | truthful_unavailable | Inspect a local database schema through a configured, read-only connection. |
222 |
| READ | default | dependency_gated | truthful_unavailable | Run a bounded read-only local SQLite SELECT, PRAGMA, or WITH...SELECT query. |
223 |
| WRITE | default | dependency_gated | service_dispatch | Read PowerPoint content or save a copy through the existing Office policy boundary. |
224 |
| READ | default | dependency_gated | service_dispatch | Read Outlook folder and message headers through the existing Office policy boundary. |
225 |
| READ | default | dependency_gated | truthful_unavailable | Extract bounded PDF text and tables through a local document provider. |
226 |
| WRITE | default | dependency_gated | service_dispatch | Create a deterministic DOCX merge plan and write only after approval. |
227 |
| READ | default | operational | service_dispatch | Propose safe, deterministic, reversible recovery steps without applying mutations. |
228 |
| DANGEROUS | default | dependency_gated | service_dispatch | Apply a current reversible recovery plan without automatic destructive retries; standard mode requires confirmation and trusted Full Bypass skips jarvis approval. |
229 |
| WRITE | no | feature_disabled | truthful_unavailable | Import a compatible skill descriptor after validation and permission review. |
230 |
| EXECUTE | no | planned | truthful_unavailable | Plan bounded parallel subagents with ownership, collision, approval, and cancellation metadata. |
231 |
| READ | default | operational | service_dispatch | Search JARVIS memory using SQLite FTS5, optional vector similarity, and optional graph expansion within the requested project scope. |
232 |
| WRITE | default | operational | service_dispatch | Persist a durable JARVIS memory record in the canonical SQLite store. |
233 |
| WRITE | default | operational | service_dispatch | Create or update a relationship between two canonical memory records. |
234 |
| READ | default | operational | service_dispatch | Return canonical memory counts for global memory or the active project scope. |
235 |
| READ | default | operational | service_dispatch | Build standardized Context Intelligence 2.0 packet: project, Git, ranked workspace context, memory, and goal context under the active workspace boundary. |
236 |
| EXECUTE | default | operational | service_dispatch | Run the authoritative verification gates: typecheck, tests, build, runtime health, and Git diff review. Returns VERIFIED, PARTIALLY_VERIFIED, FAILED, or BLOCKED evidence. |
237 |
| READ | default | operational | deterministic_operation | Tool Router 2.0: rank the best MCP tools for a natural-language intent (typo-tolerant) with required/optional arguments, so the correct tool is invoked accurately on the first attempt. |
238 |
| READ | default | operational | deterministic_operation | Recovery Engine 2.0: classify a failure from verification/process/tunnel/Git evidence and recommend ordered recovery actions, including bounded automatic retry when the failure is transient. |
239 |
| EXECUTE | default | operational | service_dispatch | Execute multiple MCP tools with parallel, dependency-aware, timeout, cancellation, and partial-result handling. |
Security and operational model
The runtime follows fail-closed ownership and mutation boundaries. Process and tunnel ownership are verified before stale recovery, destructive administrative actions require the configured approval path, and recovery actions preserve checkpoint/provenance evidence. Runtime diagnostics are UTF-8 and bounded; secrets are kept outside ordinary Web API responses.
Legacy jarvis
The original jarvis Desktop runtime and its Windows packaging remain available for the transition period. JARVIS development does not require rebuilding or packaging the legacy Desktop EXE.
For ChatGPT web and other supported OpenAI surfaces, the legacy jarvis Desktop still uses the official OpenAI Secure MCP Tunnel. The tunnel is outbound-only: tunnel-client runs beside jarvis, reaches OpenAI over outbound HTTPS, and forwards MCP work to jarvis's Desktop loopback HTTP MCP endpoint (shown in the Desktop dashboard) without opening a public inbound port on the Windows machine. The tunnel profile is initialized against that loopback endpoint with the official sample profile:
tunnel-client.exe init --force --sample sample_mcp_remote_no_auth --profile jarvis --tunnel-id '<tunnel-id>' --control-plane-api-key-ref 'env:CONTROL_PLANE_API_KEY' --health-listen-addr '127.0.0.1:0' --mcp-server-url '<desktop-loopback-http-mcp-url>'License
MIT
This server cannot be installed
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 Servers
- AlicenseBqualityDmaintenanceHigh-performance MCP server giving AI agents advanced filesystem and automation capabilities on Windows, with 26 tools across file I/O, search, Git, process management, and more.262MIT
- FlicenseNot gradedqualityDmaintenanceA lightweight MCP HTTP server giving AI assistants real tools to interact with your Windows machine, including running commands, file access, system info, web search, and browser automation.1
- FlicenseNot gradedqualityCmaintenanceAn all-in-one MCP server that enables AI agents with local system capabilities like filesystem navigation, command execution, desktop automation, and a continuity memory suite for persistent cross-session project context.
- AlicenseNot gradedqualityBmaintenanceRemote-first, model-independent Windows execution MCP for ChatGPT, Claude, Gemini, and other MCP-compatible AI clients.MIT
Related MCP Connectors
Self-hosted MCP gateway: turn any API, database or MCP server into AI connectors — no code.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Private-by-default, local-first memory/context/task orchestrator for MCP apps and agents.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/skyliner2008/JARVIS-MCP'
If you have feedback or need assistance with the MCP directory API, please join our Discord server