mefcs-mcp
Click on "Deploy 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., "@mefcs-mcp switch to preset 5 and show me its amp, cab, and gain settings"
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.
mefcs-mcp
An MCP server that lets AI assistants (Claude, Cursor, VS Code Copilot, Gemini CLI, Codex and any other MCP client) control an M-VAVE Tank-G guitar pedal and pull tones from TONE3000.
It talks to the pedal over USB with the same protocol as the official M-EFCS app. The protocol is reverse engineered and documented in docs/PROTOCOL.md. It uses the USB helper that ships with M-EFCS, so nothing has to be installed on the pedal.
Unofficial project, not affiliated with M-VAVE / Sinco or TONE3000. Firmware update, repair and factory-restore functions are intentionally not implemented.
Contents
Related MCP server: Pocket Master MCP
What it can do
Pedal (Tank-G)
Read all 36 presets, or the live values of the active preset.
Switch presets, toggle bypass.
Edit any value (amp slot, gain, EQ, volume, gate, MOD / DELAY / REVERB type and knobs, cab slot, LED color) and save it to the pedal. A backup is taken automatically first.
Tweak the active preset live without saving.
Export and import
.atkgpreset files, the same format as M-EFCS "Share/Import all presets". Imports are backed up and verified.Upload an amp capture (NAM A2
.nam, converted to "A2 Lite (Original)" like M-EFCS does) to amp slot 1-9, or preview it.Upload a cab impulse response (
.wav, any rate/bit depth, resampled to 44.1 kHz) to cab slot 1-8, or preview it.
TONE3000
Search tones (NAM A2 captures by default, or IRs) by text, gear type, make/model, tags or creator.
See a tone's details and its downloadable models.
Download
.nam/.wavfiles.Install a TONE3000 model straight onto a pedal slot in one step.
When keys or a login are missing, the server opens the right TONE3000 page in your browser for you.
Requirements
OS | Windows 10/11. The USB helper is a Windows program. |
Node.js | 20 or newer ( |
M-EFCS | Installed. The server uses its |
Pedal | M-VAVE Tank-G (v2) connected by USB |
TONE3000 | Free account, only for the TONE3000 tools |
Important: only one program can talk to the pedal at a time. Close M-EFCS before the AI uses a pedal tool. Ask the AI to run tankg_disconnect before opening M-EFCS again.
Installation
# 1. Get the code
git clone https://github.com/cayocan/mefcs-mcp.git
cd mefcs-mcp
# 2. Install dependencies and build
npm install
npm run build
# 3. Optional: run the tests (they compare the output with captured M-EFCS traffic)
npm testThe server entry point is dist/src/server.js. Once the package is published on npm you can skip the clone and use npx -y mefcs-mcp instead (see Publishing).
Find the full path of mefcs_ks_bridge.exe inside your M-EFCS folder (for example C:\Program Files\M-EFCS\mefcs_ks_bridge.exe). Standard locations are detected automatically; otherwise set MEFCS_BRIDGE_PATH.
Connecting it to an AI client
MCP is an open standard, so the same server works in every client that supports local (stdio) MCP servers. In each example, replace the paths with yours.
Claude Desktop
Edit %APPDATA%\Claude\claude_desktop_config.json and restart Claude Desktop:
{
"mcpServers": {
"mefcs": {
"command": "node",
"args": ["C:\\path\\to\\mefcs-mcp\\dist\\src\\server.js"],
"env": {
"MEFCS_BRIDGE_PATH": "C:\\Program Files\\M-EFCS\\mefcs_ks_bridge.exe",
"MEFCS_BACKUP_DIR": "C:\\path\\to\\mefcs-mcp\\backups",
"TONE3000_DOWNLOAD_DIR": "C:\\path\\to\\mefcs-mcp\\downloads"
}
}
}
}Claude Code
claude mcp add mefcs --scope user `
-e MEFCS_BRIDGE_PATH="C:\Program Files\M-EFCS\mefcs_ks_bridge.exe" `
-- node "C:\path\to\mefcs-mcp\dist\src\server.js"Cursor
%USERPROFILE%\.cursor\mcp.json (global) or .cursor/mcp.json (project) takes the same mcpServers block as Claude Desktop.
VS Code (GitHub Copilot agent mode)
.vscode/mcp.json:
{
"servers": {
"mefcs": {
"type": "stdio",
"command": "node",
"args": ["C:\\path\\to\\mefcs-mcp\\dist\\src\\server.js"],
"env": { "MEFCS_BRIDGE_PATH": "C:\\Program Files\\M-EFCS\\mefcs_ks_bridge.exe" }
}
}
}Windsurf, Cline, Gemini CLI and others
They all accept the same three things: command = node, args = path to dist/src/server.js, and env = the variables below. The mcpServers JSON block from Claude Desktop goes in:
~/.codeium/windsurf/mcp_config.json(Windsurf)Cline's MCP settings
~/.gemini/settings.json(Gemini CLI)
OpenAI Codex CLI
%USERPROFILE%\.codex\config.toml:
[mcp_servers.mefcs]
command = "node"
args = ["C:\\path\\to\\mefcs-mcp\\dist\\src\\server.js"]
env = { MEFCS_BRIDGE_PATH = "C:\\Program Files\\M-EFCS\\mefcs_ks_bridge.exe" }Clients that only accept remote (HTTPS) MCP servers, such as web chat apps, cannot reach a USB pedal directly. The server has to run on the PC that has the pedal. See Publishing.
Configuration
Variable | Default | Meaning |
| auto-detected in Program Files / LocalAppData | Full path to |
|
| Automatic preset backups ( |
|
| Downloaded |
| (none) | TONE3000 publishable key; can also be saved at runtime |
| (none) | TONE3000 secret key |
|
| Saved TONE3000 session |
|
| Keys saved with |
| (unset) | Set to |
TONE3000 setup
You only do this once, and you never need to edit config files for it:
Ask the AI to search TONE3000 (for example "search TONE3000 for a Fender Twin"). Without keys, the server opens https://www.tone3000.com/settings in your browser.
Log in there and generate API keys. Give the publishable key (client id) to the AI, which saves it with
tone3000_set_credentials. It is stored in~/.mefcs-mcp/config.jsonon your PC.The server then opens the TONE3000 approval page with your code. Click approve, and the AI runs
tone3000_login_finish.Done. The session is saved and refreshed automatically. From now on, searches and downloads just work.
Prefer not to log in? Put the secret key (t3k_cs_…) in TONE3000_API_KEY yourself instead. Don't paste secret keys into a chat.
Notes:
The default rate limit is 100 requests per minute.
Searches default to NAM A2 captures, because the Tank-G loads the lite part of A2 files. Standard (A1) WaveNet captures cannot be converted.
TONE3000's API terms ask independent developers to credit TONE3000.
Tool reference
Pedal
Tool | Inputs | What it does |
| — | Opens the USB session (M-EFCS must be closed). Other tools connect on demand. |
| — | Releases the pedal so M-EFCS can be used. |
| — | Active preset ( |
| — | All 36 stored presets, decoded. |
|
| One stored preset, or the live values of the active one. |
|
| Switch preset. |
|
| Edit and save. Backs up first. Also applied live if the preset is active. |
|
| Change the active preset without saving. |
|
| Global bypass. |
|
| Save all presets to an |
|
| Replace all presets with an |
|
| Convert a NAM A2 capture and store it in an amp slot. Without |
|
| Store a WAV IR in a cab slot (max ~30 ms at 44.1 kHz). Without |
|
| Decode an |
| — | Effect type indexes. |
changes fields (all optional):
Field | Range |
| bool |
| 1-9 |
| 0-100 |
| 0-100, 0 = off |
|
|
| 0-8, 0 = off |
|
|
Effect types follow the knob zones in the official manual:
Block | type 0 | type 1 | type 2 | amount | param2 |
MOD | chorus | phaser | tremolo | mix | rate |
DELAY | analog | tape | dual | mix | time |
REVERB | room | spring | cloud | decay | mix |
amount ranges from 0 to 42 (0 to 40 for type 2).
TONE3000
Tool | Inputs | What it does |
| — | Shows the auth mode and whether you are logged in. |
|
| Saves keys locally and starts the login (opens the approval page). |
| — | Device login. Start opens the browser; finish waits up to 60 s for approval. |
|
| Search tones. |
|
| Details and downloadable models. |
|
| Download a |
|
| Download and send to an amp slot ( |
Example prompts
"Which preset is active on my Tank-G, and what does it have?"
"Turn off the reverb on 5D and lower the gain to 60."
"Make 1A's chorus subtler without saving, I want to hear it first."
"Back up all my presets to D:\Backups\tankg.atkg."
"Search TONE3000 for a Vox AC30 capture with lots of downloads and preview the best one."
"Install the Marshall JCM800 capture you found on amp slot 4 and point preset 3C at it."
"Find a Celestion V30 4x12 IR on TONE3000 and put it on cab slot 5."
Safety
There is no code path to firmware update, repair or factory restore, and every write is checked against an address allow-list (see
assertAllowedWrite).Preset edits and imports back up all 36 presets first. Imports are verified by reading the bank back.
Values are validated: knobs 0-100 (the range M-EFCS uses), slots 1-9 / 1-8, effect amounts within the type's zone. Untouched legacy values are kept as they are.
Slot uploads overwrite the slot. Without
slotthey only preview, and nothing is stored.The browser is only opened for
https://www.tone3000.compages.TONE3000 keys and tokens stay on your PC (
~/.mefcs-mcp).
How it works
AI client ──MCP (stdio)──> mefcs-mcp ──stdin/stdout──> mefcs_ks_bridge.exe ──USB MIDI (SysEx)──> Tank-G
└──HTTPS──> TONE3000 APIsrc/protocol/bridge.ts: spawns the M-EFCS helper (open-raw) and speaks its text protocol (send <hex>→data <hex>).src/protocol/framing.ts: SysEx frames: 8-bit frame → 7-bit LSB-first bitstream, inverted-sum checksum.src/devices/tankg.ts: memory map (preset bank, live preset, select/bypass/commit, model buffers, slot store) plus the write allow-list.src/devices/tankg-presets.ts: 21-byte preset record codec (.atkglayout).src/codecs/a2lite.ts: NAM A2 → device amp buffer. It takes the lite submodel, transposes the convolution weights and precomputes the zero-input state.src/codecs/ir.ts: WAV → device cab buffer (44.1 kHz, int24, max 1358 samples).src/tone3000/client.ts: TONE3000 API v1 (device login, search, models, download).
Troubleshooting
Problem | Fix |
"M-EFCS is open" | Close M-EFCS; only one program can use the pedal. |
"mefcs_ks_bridge.exe not found" | Set |
"No M-VAVE device found" | Check the USB cable/port; make sure the pedal is on. |
"Timed out waiting for the pedal" | Unplug and replug the pedal, then retry. Make sure nothing else is using it. |
M-EFCS can't find the pedal after using the AI | Ask the AI to run |
"not an A2 (SlimmableContainer) capture" | Pick an A2 capture on TONE3000 (search already filters for them). |
TONE3000 rate limit | Wait a minute (100 requests/min). |
A preset change sounds wrong | Restore the latest backup in |
Development
npm run build # TypeScript -> dist/
npm test # golden tests against captured M-EFCS traffic + TONE3000 client tests
node scripts/smoke-test.mjs out.atkg # live read-only test on a real pedal (M-EFCS closed)
node scripts/normalize-knobs.mjs --dry-runTwo golden tests compare against third-party TONE3000 captures that are not redistributed in this repo. They are skipped automatically when test/fixtures/mesa.nam / ir8.wav are absent.
To add another M-VAVE device (MK300, Tank-B, …), add a driver under src/devices/ with that device's memory map and model format. The framing and the bridge are shared. Capture its traffic first (see docs/PROTOCOL.md).
Publishing
The goal is that anyone can run it with npx -y mefcs-mcp and find it in MCP directories.
GitHub: push this repository (for example
github.com/cayocan/mefcs-mcp) and add a license.npm:
npm login, thennpm publish.prepublishOnlybuilds and tests first. After this, clients can use:{ "command": "npx", "args": ["-y", "mefcs-mcp"], "env": { "MEFCS_BRIDGE_PATH": "..." } }Official MCP Registry:
package.jsonalready has"mcpName": "io.github.cayocan/mefcs-mcp"and the repository has aserver.json. Installmcp-publisher, then runmcp-publisher login githubandmcp-publisher publish. Keep the versions inpackage.jsonandserver.jsonequal.Directories (optional): Smithery, Glama, mcp.so, and the Cursor / VS Code MCP galleries can list it from the npm package or the registry.
Because the pedal is a local USB device, this is a local (stdio) server. Remote-only clients would need an HTTP transport running on the PC with the pedal, exposed through a secure tunnel. That is not included yet.
License
MIT © Cayo Aguiar. M-VAVE, Tank-G, M-EFCS and TONE3000 are trademarks of their respective owners.
Available Tools
23 toolstankg_connectConnect to the Tank-GA
Open the USB bridge to the pedal. M-EFCS must be closed.
Inputs: no named parameters.
tankg_disconnectDisconnectB
Release the pedal so M-EFCS can be opened again.
Inputs: no named parameters.
tankg_effect_typesEffect typesB
Effect type indexes and names (0 = first knob zone, 2 = last).
Inputs: no named parameters.
tankg_export_presetsExport presetsB
Save all 36 presets to an .atkg file (same format as M-EFCS 'Share all presets').
Inputs: path (required).
tankg_get_presetGet presetB
Read one stored preset, or the live (unsaved) values of the active one.
Inputs: live, preset.
tankg_import_presetsImport presetsA
Replace all 36 presets with an .atkg file. A backup of the current presets is saved first.
Inputs: path (required).
tankg_install_tone3000_modelInstall a TONE3000 model on the Tank-GB
Download a TONE3000 model and send it to the pedal: .nam (A2) goes to amp slot 1-9, .wav IR goes to cab slot 1-8. Without slot it is only previewed. Overwrites the slot.
Inputs: modelId (required), name, slot, cabLevel.
tankg_list_presetsList presetsA
Read all 36 stored presets from the pedal.
Inputs: no named parameters.
tankg_read_preset_fileRead .atkg fileA
Decode an .atkg preset file offline (no pedal needed).
Inputs: path (required).
tankg_select_presetSelect presetC
Make a preset active on the pedal.
Inputs: preset (required).
tankg_set_bypassBypassC
Turn global bypass on or off.
Inputs: on (required).
tankg_set_live_valuesTweak active preset (unsaved)C
Change the active preset's live values without saving.
Inputs: changes (required).
tankg_statusPedal statusC
Active preset and bypass state.
Inputs: no named parameters.
tankg_update_presetUpdate presetA
Change values of a stored preset. With save=true (default) it is written to the pedal's flash (a backup of all presets is taken first). If the preset is active, the change is also applied live.
Inputs: preset (required), changes (required), save.
tankg_upload_amp_captureUpload amp captureA
Convert a TONE3000/NAM A2 capture (.nam) to 'A2 Lite (Original)' and send it to the pedal. With slot 1-9 it overwrites that amp slot; without slot it is only previewed (not stored).
Inputs: namPath (required), name, slot.
tankg_upload_cab_irUpload cab IRA
Send a WAV impulse response to the pedal (resampled to 44.1 kHz, max ~30 ms). With slot 1-8 it overwrites that cab slot; without slot it is only previewed.
Inputs: wavPath (required), slot, level.
tone3000_auth_statusTONE3000 login statusC
Show how the MCP authenticates with TONE3000.
Inputs: no named parameters.
tone3000_download_modelDownload from TONE3000C
Download one model (.nam) or IR (.wav) by model id.
Inputs: modelId (required), dir.
tone3000_get_toneTONE3000 tone detailsC
Tone details plus its downloadable models (A2 NAM models or IRs).
Inputs: toneId (required).
tone3000_login_finishTONE3000 login (finish)A
Wait for the user to approve the device code (up to 60 s per call).
Inputs: no named parameters.
tone3000_login_startTONE3000 login (start)A
Start the TONE3000 device login. Returns a code and a link; the user approves it in the browser, then call tone3000_login_finish. Needs TONE3000_CLIENT_ID.
Inputs: no named parameters.
tone3000_searchSearch TONE3000B
Search TONE3000 tones. Defaults to NAM A2 captures (what the Tank-G can load). Use format='ir' for cab IRs. gears: amp, amp-cab, pedal, outboard, cab, space, experimental.
Inputs: page, sort, tags, gears, +7 more.
tone3000_set_credentialsSave TONE3000 keysA
Save the TONE3000 publishable key (client id, preferred) and/or secret key (t3k_cs_...) on this computer, then start the login. Keys come from https://www.tone3000.com/settings.
Inputs: apiKey, clientId.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 23 tools
Most tools are clearly separated by resource and action, and the tankg_/tone3000_ prefixes help distinguish pedal control from cloud operations. There is some overlap between tankg_install_tone3000_model, tankg_upload_amp_capture, and tankg_upload_cab_ir because all can send captures or IRs to the pedal, though the descriptions make their intended use mostly recoverable.
Names follow a consistent snake_case convention with predictable domain prefixes (tankg_ and tone3000_) and verb/noun phrasing such as list_presets, get_preset, select_preset, and update_preset. The few state-related names like tankg_status and tankg_connect still fit the same overall pattern.
The server exposes 23 tools, which is on the heavy side for the apparent scope. The surface combines pedal control, preset file handling, and TONE3000 authentication/downloading, so the size is explainable but borderline rather than well-scoped.
The tool set covers the main lifecycle: connecting, reading/writing presets, live edits, bypass, import/export, local file reading, uploads, and TONE3000 search/auth/download/install. Minor gaps remain, such as no explicit create_preset or delete_preset operation, though import/update workflows may allow agents to work around some of these cases.
Maintenance
Related MCP Connectors
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
Image and video AI tools and your own pipelines, run from any AI assistant.
Browser DAW for AI assistants: build multitrack songs with MIDI, effects, stems and generation.
- DNS-DATA RemoteDeprecatedcom.dns-data
Connect AI assistants to authorized Windows and macOS devices for files, commands, and management.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceLets AI assistants control a NUX MG-30 guitar processor over USB MIDI by picking amp, cab, drive, modulation, delay and reverb models, toggling blocks, and setting knobs by name. It can also read, save, rename, back up and restore presets on the device, and apply whole tone recipes from JSON patch files.57 PyPI1MIT
- AlicenseNot gradedqualityBmaintenanceEnables an AI assistant to control a Sonicake Pocket Master guitar processor over USB MIDI by describing tones in plain words — selecting amp/cab/effect models, setting parameters, reordering the signal chain, and saving, renaming, backing up or restoring presets. It also provides offline tools for inspecting firmware and rebuilding it with a translated UI, such as Russian.58 PyPI2MIT
- AlicenseNot gradedqualityBmaintenanceEnables users to build and tweak BOSS GT-1 guitar tones and patches through plain-language conversation, with Claude writing amp, drive, and effect settings over USB. It can also listen to the pedal's audio, adjust levels and noise suppression, manage user slots, and run a metronome.MIT
- AlicenseAqualityBmaintenanceEnables AI coding agents to shape BOSS KATANA Gen 3 amplifier tones from natural-language descriptions, moving the knobs live inside BOSS TONE STUDIO over USB. Agents can read the current patch, set amp and effect parameters in one call, and let changes play through the amp instantly.69MIT