agents-md-generator

MCP server that analyzes codebases with tree-sitter and generates AGENTS.md files.

Compatible with any MCP-capable client: Claude Code, Gemini CLI, Cursor, Windsurf, and others.

How it works: The server does all the heavy lifting locally — AST parsing, incremental change detection, environment variable scanning, entry point detection. It writes a compact structured payload to disk and returns step-by-step instructions to your AI client. The client reads the payload and writes AGENTS.md. No large data travels over the MCP wire.

Supported Languages

Python · C# · TypeScript · JavaScript · Go

Installation

See INSTALLATION.md for the full guide including prerequisites and troubleshooting.

Requirements: Python 3.11+, uv, Git, and any MCP-compatible client.

Claude Code

claude mcp add agents-md uvx agents-md-generator

Or add it manually to ~/.claude.json (Linux/macOS) or %USERPROFILE%\.claude.json (Windows):

{
  "mcpServers": {
    "agents-md": {
      "command": "uvx",
      "args": ["agents-md-generator"]
    }
  }
}

Gemini CLI

Add it to ~/.gemini/settings.json:

{
  "mcpServers": {
    "agents-md": {
      "command": "uvx",
      "args": ["agents-md-generator"]
    }
  }
}

Other MCP clients (Cursor, Windsurf, etc.)

The server uses stdio transport. Add this entry to your client's MCP config under mcpServers:

"agents-md": {
  "command": "uvx",
  "args": ["agents-md-generator"]
}

Restart your client — uvx downloads the package automatically on first run.

Usage

Once registered, ask your AI client:

"Generate the AGENTS.md for this project"

The client will call generate_agents_md automatically.

Tool Parameters

Note on force_full_scan: Use this only when explicitly requested. When asking Claude to improve or update an existing AGENTS.md, leave it as false — the incremental scan already provides all the data needed.

What Gets Generated

The generated AGENTS.md follows the agents.md open standard. It is written as a README for AI agents, not as documentation for humans. Sections include:

• Project Overview — tech stack and top-level architecture shape
• Architecture & Data Flow — detected layers or domains with data flow direction
• Conventions & Patterns — naming rules, export contracts, import rules, and how to add new entities end-to-end
• Environment Variables — variables detected in source files and .env.example
• Setup Commands — exact install and run commands from package.json, Makefile, etc.
• Development Workflow — build, watch, and dev server commands
• Testing Instructions — test commands and framework info (if detected)
• Code Style — lint/format commands (if config files detected)
• Build and Deployment — CI pipeline info (if detected)

Sections with no detected data are omitted entirely.

How Incremental Scanning Works

1. First run (cold start): All git-tracked source files are parsed with tree-sitter and cached
2. Subsequent runs: Only files whose SHA-256 hash changed since the last scan are re-parsed
3. Semantic diff: For modified files, only changed public symbols are included in the payload
4. No source changes? The tool stops and asks whether you want to improve the existing AGENTS.md content anyway
5. Private symbols and test file internals are excluded from both cache and payload — only the public API surface matters for AGENTS.md

How Large Payloads Are Streamed

For large codebases the analysis payload can be too big to return inline over the MCP wire. The server handles this transparently through a second tool: get_payload_chunk.

Flow:

1. generate_agents_md runs the full analysis, writes the payload to disk, and returns a small response with total_chunks and instructions
2. The client calls get_payload_chunk(project_path, chunk_index=0), then increments chunk_index until the response contains has_more: false
3. The client concatenates all data fields in order and parses the result as JSON
4. The payload file is automatically deleted after the last chunk is read

This flow is pure MCP — no filesystem access required from the client side. Any MCP-compatible client can follow it.

Cache and Payload Location

All runtime artifacts are stored outside your project, in the user cache directory:

The <project-hash> is a SHA-256 of the project's absolute path — unique per project. Nothing is written to your repository.

Note: The server also writes a temporary payload.json to this directory during analysis, but it is managed entirely by the get_payload_chunk tool and deleted automatically after the last chunk is read. You never need to access it directly.

Project Configuration

Create .agents-config.json at your project root to customize behavior. This file is optional — all fields have defaults.

{
  "impact_threshold": "medium",
  "exclude": [
    "**/node_modules/**",
    "**/bin/**",
    "**/obj/**",
    "**/.git/**",
    "**/dist/**",
    "**/build/**",
    "**/__pycache__/**",
    "**/*.min.js",
    "**/*.min.css",
    "**/*.bundle.js",
    "**/vendor/**",
    "**/packages/**",
    "**/.venv/**",
    "**/venv/**",
    "**/bower_components/**",
    "**/app/lib/**",
    "**/wwwroot/lib/**",
    "**/wwwroot/libs/**",
    "**/static/vendor/**",
    "**/public/vendor/**",
    "**/assets/vendor/**",
    "**/site-packages/**"
  ],
  "include": [],
  "languages": "auto",
  "agents_md_path": "./AGENTS.md",
  "max_file_size_bytes": 1048576,
  "dir_aggregation_threshold": 8
}

Options

You can commit .agents-config.json to share exclusion rules and thresholds with your team.

Impact Threshold

The impact_threshold controls which symbol changes are included in incremental scan payloads. Changes below the threshold are silently ignored — AGENTS.md is not regenerated for them.

Choosing a threshold:

• "high" — Only regenerate AGENTS.md for breaking or structural changes. Best for large, stable codebases where minor additions are frequent.
• "medium" (default) — Regenerate when the public API surface grows or changes. Suitable for most projects.
• "low" — Regenerate on any public symbol change. Best for early-stage projects where the architecture is still evolving.

What the Analysis Detects

Environment Variables

The server scans all source files for environment variable references using language-specific patterns:

It also parses .env.example, .env.template, and .env.sample files at the project root.

Entry Points

Files named index, main, app, server, program, bootstrap, or startup (with any supported extension) are detected as entry points and annotated with their inferred role (e.g., "HTTP server bootstrap", "Electron main process").

Public API Surface

Tree-sitter parses each source file and extracts public symbols — classes, functions, methods, interfaces — filtering out private/protected members and underscore-prefixed symbols. For classes and structs, constructors (when they have parameters) and public properties are also included, revealing dependency injection patterns and data shapes. Interface methods are always included as they define the public contract. These are used to detect naming conventions, DI patterns, and export contracts across layers.

Credits

AGENTS.md format based on the open agents.md standard.