robot-md-mcp

The one-command bridge from a ROBOT.md file to any MCP-aware agent. Claude Code, Claude Desktop, Cursor, Zed, Cline, Continue.dev, OpenAI Codex CLI, Google Gemini CLI — any MCP client speaking stdio transport reads your robot's manifest as structured resources. Zero config files.

Where this fits in the stack

This repo is the agent bridge — an MCP server that speaks the open Model Context Protocol and serves your ROBOT.md to any MCP-aware planner. Everything else is independent; adopt one, or all seven.

| Layer | Piece | What it is | |---|---|---| | Declaration | ROBOT.md | The file a robot ships at its root. YAML frontmatter + markdown prose. Declares identity, capabilities, safety gates. Spec + Python CLI (init, validate, calibrate, register). | | Agent bridge ← this | robot-md-mcp | MCP server that exposes a ROBOT.md as 4 MCP resources (frontmatter, capabilities, safety, body) + 2 tools (validate, render). One claude mcp add away. | | Wire protocol | RCAN | How robots, gateways, and planners talk. Signed envelopes, LoA enforcement, PQC crypto. Think HTTP for robots. | | Python SDK | rcan-py | pip install rcan — RCANMessage, RobotURI, ConfidenceGate, HiTLGate, AuditChain. | | TypeScript SDK | rcan-ts | npm install rcan-ts — same API surface for Node + browser. | | Registry | Robot Registry Foundation | Permanent RRN identities. Public resolver at /r/<rrn>. Like ICANN for robots. | | Reference runtime | OpenCastor | Open-source robot runtime — connects LLM brains to hardware bodies. One implementation of RCAN. |

What it does

Reads a local ROBOT.md and exposes it to an MCP client as:

Resources — the client can read at will:
- robot-md://<robot_name>/frontmatter (application/json)
- robot-md://<robot_name>/capabilities (application/json)
- robot-md://<robot_name>/safety (application/json)
- robot-md://<robot_name>/body (text/markdown)
Tools — the client invokes on operator request:
- validate → { ok, summary, errors }
- render → canonical YAML of the frontmatter

The server re-reads the file on every call. No cache, no watcher, no runtime config.

Not in v0.1 — deferred to v0.2

No signature verification. ROBOT.md v0.2 will add signed manifests (.sig) and a key-binding-at-RRN-mint flow; see spec/v0.2-design.md.
No robot dispatch. invoke_skill / query_status arrive after the v0.2 signing decisions in §13 are finalized.
No multi-manifest / fleet mode.

Claude Code (CLI) — one command

The fastest way. One line, no config files:

claude mcp add robot-md -- npx -y robot-md-mcp /absolute/path/to/ROBOT.md

That's it. Open Claude Code — it now reads your robot's frontmatter, capabilities, safety block, and prose body as MCP resources, and can invoke the validate and render tools on request. No harness config, no provider setup, no YAML wrangling.

To confirm it's wired up:

claude mcp list | grep robot-md

To remove: claude mcp remove robot-md.

Install (standalone)

If you want to run the server yourself:

npx robot-md-mcp /path/to/ROBOT.md

Node 18.20+ required. The claude mcp add flow above uses npx under the hood, so no global install needed for the Claude Code path.

Claude Desktop config

Add this to your claude_desktop_config.json:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "robot-md": {
      "command": "npx",
      "args": ["-y", "robot-md-mcp", "/absolute/path/to/ROBOT.md"]
    }
  }
}

Restart Claude Desktop. Open a new chat — Claude now has the robot's frontmatter, capabilities, safety block, and prose body on tap.

Any MCP-aware agent harness

ROBOT.md is a file. robot-md-mcp is an MCP server. MCP is an open standard. Any agent tool that speaks MCP can ingest ROBOT.md through this server — not just Claude.

The underlying command to register is always the same:

npx -y robot-md-mcp /absolute/path/to/ROBOT.md

Stdio transport, no auth. How you wire it depends on the harness:

| Harness | How to add | |---|---| | Claude Code (CLI) | claude mcp add robot-md -- npx -y robot-md-mcp /path/to/ROBOT.md | | Claude Desktop | JSON snippet in claude_desktop_config.json (see above) | | OpenAI (Codex CLI, ChatGPT Desktop) | Add to the tool's MCP-server config — command npx, args ["-y","robot-md-mcp","/path/to/ROBOT.md"] | | Google Gemini CLI | Add to ~/.gemini/settings.json under mcpServers with the command above | | Cursor / Zed / Cline / Continue.dev / VS Code MCP extensions | Add via the tool's MCP settings with the same command | | Anything else speaking MCP stdio | Register the npx command; that's it |

Create and place one ROBOT.md. Any agent — any provider — reads the same file. That's the whole point.

Tier-0 adoption loop

# Generate a draft from visible hardware
pip install robot-md
robot-md autodetect --write ./ROBOT.md

# Edit the TODOs (robot name, physics type, DoF, capabilities)
# Then point Claude Desktop at it:
# -> add the JSON snippet above to claude_desktop_config.json

API surface

import {
  // Core
  parseRobotMd,
  ParseError,
  validateParsed,
  renderYaml,
  createServer,
  VERSION,
  // Types
  type ParsedRobotMd,
  type ValidateResult,
  type ServerHandle,
} from "robot-md-mcp";

All are importable for programmatic use — e.g. building a custom MCP server on top, or running the parser/validator as part of a larger TypeScript tool.

Development

git clone https://github.com/RobotRegistryFoundation/robot-md-mcp
cd robot-md-mcp
npm install
npm test            # vitest: parser, validator, render, server, bin
npm run build       # tsup → dist/
npm run sync-schema # refresh bundled schema from ../robot-md

Contributing

Schema lives at RobotRegistryFoundation/robot-md — PR changes there, then re-run npm run sync-schema here.
Small, focused PRs welcome.

License

Apache 2.0.

MCP Servers