@latent-space-labs/open-auto-doc-mcp

MCP server for open-auto-doc — exposes AI-analyzed codebase documentation to Claude Code and other MCP-compatible AI assistants.

Setup

The easiest way to set up is through the main CLI:

# After generating docs with open-auto-doc
open-auto-doc setup-mcp

This creates .mcp.json in your project root. Claude Code automatically discovers the tools next time you open the project.

Manual .mcp.json setup

{
  "mcpServers": {
    "project-docs": {
      "command": "npx",
      "args": ["-y", "@latent-space-labs/open-auto-doc-mcp", "--project-dir", "."]
    }
  }
}

Available Tools

| Tool | Description | |---|---| | get_project_overview | Project purpose, tech stack, summary stats | | search_documentation | Full-text search across all documentation | | get_api_endpoints | API endpoints with params, auth, request/response | | get_components | UI components with props and usage examples | | get_data_models | Data models with fields, types, relationships | | get_architecture | Modules, data flow, patterns, entry points | | get_diagram | Mermaid diagrams (architecture, ER, flow) | | get_business_rules | Domain concepts, rules, workflows |

All tools accept an optional repo parameter for multi-repo setups.

Available Resources

| URI | Description | |---|---| | docs://overview | Project overview | | docs://architecture | Architecture details | | docs://getting-started | Getting started guide | | docs://diagrams/{diagramId} | Individual Mermaid diagrams |

CLI Options

open-auto-doc-mcp [options]

Options:
  --project-dir <path>   Project directory containing .autodocrc.json
  --cache-dir <path>     Direct path to .autodoc-cache directory

If no options are given, the server searches the current directory for .autodocrc.json.

How It Works

When you run open-auto-doc init or generate, the AI analysis results are saved as JSON cache files in <outputDir>/.autodoc-cache/. Each analyzed repo gets a <repo-name>-analysis.json file containing the full structured analysis.

The MCP server reads these cache files at startup, loads the analysis results into memory, and exposes them through the Model Context Protocol via stdio transport.

┌──────────────────┐      ┌───────────────────┐      ┌──────────────────┐
│  open-auto-doc   │      │   .autodoc-cache/  │      │  MCP Server      │
│  init / generate │─────▶│   *-analysis.json  │◀─────│  (stdio)         │
│                  │ save │                    │ read │                  │
└──────────────────┘      └───────────────────┘      └────────┬─────────┘
                                                              │
                                                     MCP protocol (stdio)
                                                              │
                                                     ┌────────▼─────────┐
                                                     │  Claude Code     │
                                                     │  or any MCP      │
                                                     │  client          │
                                                     └──────────────────┘

Key design decisions:

• Separate package — no dependency on the analyzer or generator, so npx startup is fast (~2 seconds) and install size is small
• Read-only — the server only reads cached JSON, it never modifies files or runs analysis
• Stdio transport — standard for Claude Code MCP servers, launched as a subprocess
• Multi-repo — loads all cached analysis files; tools accept a repo parameter to target a specific repo

Cache discovery order

1. --cache-dir <path> — direct path to the .autodoc-cache directory
2. --project-dir <path> — scans for .autodocrc.json → reads outputDir → finds cache
3. Default — scans CWD and docs-site/ for .autodocrc.json, or looks for .autodoc-cache/ directly

What's in the cache

Each *-analysis.json file contains:

• Architecture — project purpose, tech stack, modules, data flow, patterns, Mermaid diagrams
• API endpoints — method, path, parameters, request/response bodies, auth requirements
• Components — name, description, props, usage examples, file paths
• Data models — fields, types, constraints, relationships
• Business logic — domain concepts, rules, workflows, invariants
• Getting started — prerequisites, installation, quick start, configuration
• Features — feature list, use cases, categories
• Configuration — config items, env vars, config files
• Error handling — error codes, common errors, debugging tips

License

MIT