docmd includes a native Model Context Protocol (MCP) server, enabling AI development agents to interact with your documentation workspace programmatically over a secure, local connection.
What is MCP?
The Model Context Protocol is an open standard for connecting AI models to external tools and data sources. It uses JSON-RPC 2.0 messages over a transport layer (stdio, HTTP). docmd implements the stdio transport — the agent spawns docmd mcp as a child process and communicates via stdin/stdout.
Quick Start
docmd mcp
This starts the MCP server over stdio. No network ports are opened — all communication happens through standard input/output streams.
Claude Desktop Configuration
Add to your Claude Desktop claude_desktop_config.json:
{
"mcpServers": {
"docmd": {
"command": "npx",
"args": ["@docmd/core", "mcp"],
"cwd": "/path/to/your/docs/project"
}
}
}
Cursor / Windsurf Configuration
Add to your editor’s MCP settings:
{
"command": "npx @docmd/core mcp",
"transport": "stdio"
}
Available Tools
The MCP server exposes four tools that agents can call:
| Tool | Description |
|---|---|
search_docs |
Full-text search across all documentation files. Returns matching lines with file paths and line numbers. |
read_doc |
Read the raw markdown content of any documentation file by its relative path. |
validate_docs |
Run link validation across all markdown files. Returns a list of broken links with file, line, and target. |
get_llms_context |
Retrieve the complete llms-full.txt context — the unified content of the entire documentation site, optimised for LLM ingestion. |
Tool Schemas
search_docs
{
"name": "search_docs",
"inputSchema": {
"type": "object",
"properties": {
"query": { "type": "string", "description": "The term or phrase to search for." }
},
"required": ["query"]
}
}
read_doc
{
"name": "read_doc",
"inputSchema": {
"type": "object",
"properties": {
"route": { "type": "string", "description": "Relative path to the markdown file (e.g. docs/getting-started.md)." }
},
"required": ["route"]
}
}
validate_docs / get_llms_context
No input parameters required.
Protocol Details
docmd implements the MCP specification (protocol version 2025-03-26):
- Transport:
stdio— JSON-RPC 2.0 messages over stdin/stdout, one per line - Diagnostics: Logged to
stderr(does not interfere with the JSON-RPC stream) - Lifecycle:
initialize→notifications/initialized→ tool calls - Ping: Responds to
pingrequests with{}(required for connection health checks) - Capabilities: Declares
tools,resources, andprompts(tools are the primary interface)
Privacy & Security
- Local only: The server runs as a child process — no network exposure, no ports opened
- Sandboxed: File operations are restricted to the project working directory
- No telemetry: No data is sent anywhere — all processing happens on your machine
Complementary Features
The MCP server works alongside other AI-first features in docmd:
llms.txt/llms-full.txt: Generated at build time by thellmsplugin. Any agent can fetch these from your deployed site without MCP.- Copy Context widget: Browser UI button that copies page content optimised for pasting into AI chat windows.
- SKILL.md: Agent instruction manual auto-generated by
docmd init. Points to the docmd-skills knowledge base.
Use MCP when an agent needs to search, read specific files, or validate links interactively during development. Use llms-full.txt when an agent needs the entire documentation context in a single fetch (e.g., for RAG or pre-prompting).
