rtfmbro-mcp

rtfmbro provides always-up-to-date, version-specific package documentation as context for coding agents. An alternative to context7.

https://github.com/user-attachments/assets/dbe0b3b4-a42c-4e91-8bcd-a94d430ef0b8

Demonstration of rtfmbro in use, fetching a specific version (3.1.1) of Flask's readme and docs.

Overview

rtfmbro is a Model Context Protocol (MCP) server that provides real-time, version-aware documentation fetching for packages across multiple ecosystems. It bridges the gap between AI agents and accurate, up-to-date package documentation by fetching docs directly from GitHub repositories at the exact version your project uses.

Supported Ecosystems

| Ecosystem | Registry | Status | |-----------|----------|--------| | Python | PyPI | ✅ Full Support | | Node.js | npm | ✅ Full Support | | Swift | SPM | 🚧 Alpha | | GitHub | Direct | ⚠️ Fallback |

Why rtfmbro?

The Problem

• Stale Documentation: AI models often rely on outdated training data about packages
• Missing Context: Source code in node_modules etc. lacks high-level documentation, browsing it is usually token-consuming and inefficient
• Version Mismatches: Generic documentation doesn't match your specific package version, especially for legacy projects or brand new packages

The Solution

rtfmbro tries to solve these issues by:

1. Version-Precise Fetching: Retrieves documentation from the exact git tag/commit that matches your lockfile
2. Comprehensive Coverage: Extracts all documentation files (.md, .mdx, .txt, .rst, .html) from the repository
3. Intelligent Caching: SHA-based currency checking ensures docs stay fresh without unnecessary re-fetching
4. Agent Integration: Seamlessly integrates with AI coding assistants via the Model Context Protocol
5. Great DX: Zero seting up for developers, just add the server to your MCP configuration, instructions and start fetching docs

MCP Tools

The server exposes four primary tools to AI agents:

| Tool | Purpose | Parameters | Returns | |------|---------|------------|---------| | get_readme | Fetches and returns the README file for a specific package version | package, version, ecosystem | README content as string | | get_documentation_tree | Generates a comprehensive folder structure of all documentation files | package, version, ecosystem | Tree structure as string | | read_files | Reads specific documentation files with optional line range slicing | package, version, ecosystem, requests[] | Dictionary mapping paths to content | | search_github_repositories | Searches for GitHub repositories using the GitHub Search API | query, sort, order, per_page | Formatted repository search results |

Installation & Setup

Quick Start

Add the remote server to your MCP configuration:

{
  "rtfmbro": {
    "type": "http", 
    "url": "https://rtfmbro.smolosoft.dev/mcp/"
  }
}

Agent Integration

Add this policy to your agent instructions (e.g., .github/copilot-instructions.md):

## Package Documentation Policy

* Before using any package, module, or library, ALWAYS consult its documentation — your internal knowledge may be outdated and could cause errors or failures.
* Don't assume anything about the package's functionality, usage, or configuration without verifying it against the official documentation.
* If you are unsure about a user's request that may involve a package, module, or library, **ALWAYS** refer to the documentation first or ask the user to provide the package name and version.

### Workflow
1. **Identify** the package name and EXACT pinned version (e.g., "==1.0.5") from the project's lock file (uv.lock, package-lock.json, etc.)
2. **Fetch documentation** using the available tools:
   - `get_readme` - Get package README
   - `get_documentation_tree` - Browse available docs
   - `read_files` - Read specific documentation files
   - `search_github_repositories` - Search for Github packages and repositories by topic, language, or keywords
3. **Review** the documentation to understand:
   - Core functionality and API surface
   - Usage patterns and best practices  
   - Breaking changes and migration guides
   - Configuration options and defaults
4. **Apply** documentation insights to provide accurate, version-specific guidance

**Ecosystem Selection:**
- Use `pypi` for Python packages
- Use `npm` for Node.js packages  
- Use `spm` for Swift packages
- Use `gh` only as a fallback for languages not supported by the above registries (package name must be in 'owner/repo' format) or if you're unsure about the package's ecosystem. Also use `gh` when you don't find the package in the lock file.

How It Works

Documentation Workflow

1. Registry Lookup: Queries the package registry (PyPI, npm, etc.) for metadata
2. GitHub Discovery: Extracts the GitHub repository URL from package metadata
3. Version Resolution: Fetches available git tags and matches them against your semantic version
4. Smart Fetching: Clones the repository at the exact matched tag/commit
5. Content Filtering: Extracts only documentation files, removing source code and build artifacts
6. Caching & Currency: Stores results with SHA-based currency checking for efficient re-access

Caching Strategy

• SHA-Based Validation: Compares current repository commit SHA with cached version
• Automatic Invalidation: Re-fetches documentation when new commits are detected
• Persistent Storage: Maintains local cache to avoid redundant GitHub API calls
• Metadata Preservation: Stores documentation tree structure for fast browsing

Prerequisites

• Package must be published to a supported registry (PyPI, npm)
• Package metadata must contain a valid GitHub repository link
• Repository must use git tags for version management
• Documentation files must be present in the repository (not just generated sites)

Roadmap

Near Term

• [x] Ecosystem independent fallback: Implement a fallback mechanism for unsupported ecosystems
• [ ] Provide rtfmbro source code: Open source the server codebase
• [ ] Public docker image: Create a public Docker image for easy deployment
• [ ] Private repo support: Allow authenticated access to private repositories
• [ ] Add Tests: Implement unit and integration tests for core functionality
• [ ] Enhanced Python Support: Include pydocs and docstring extraction
• [ ] Search Capabilities: Search across documentation corpus

Future Ecosystems / Languages / Registries

• [ ] Rust (crates.io)
• [ ] Go (pkg.go.dev)
• [ ] Java/Kotlin (Maven Central)
• [ ] C#/.NET (NuGet)
• [ ] Ruby (RubyGems)

Source code hosting and repository support:

• [x] GitHub: Support for GitHub repositories
• [ ] Gitlab: Support for GitLab repositories
• [ ] Bitbucket: Support for Bitbucket repositories
• [ ] Launchpad: Support for Launchpad repositories

Known Issues

• Some packages may have documentation in separate standalone repos
• Large repositories may take a bit of time to clone and process initially

Similar / Additive Projects

• mcp-package-docs: Another MCP server for package documentation, focusing on documentation extraction, LSP servers, etc. May be a great supplement to rtfmbro.
• rust-docs-mcp-server: MCP server for Rust documentation, focused on Rust-specific features and documentation formats.
• mcp-ragdocs: MCP server for RAG (Retrieval-Augmented Generation) documentation, aimed at improving the documentation experience for AI models.
• godoc-mcp: MCP server for Go documentation, providing access to Go package documentation via the Model Context Protocol.
• context7: Alternative to rtfmbro

Differences between context7 and rtfmbro

|Aspect|context7|rtfmbro| |------|--------|-------| | Actuality | Scrapes documentation ahead-of-time at intransparent intervals or upon user trigger. As of writing, the "latest" Next.js docs are already 2 days old.| Fetches documentation just-in-time, ensuring it's always up-to-date. | | Version-specific docs | Theoretically allows scraping older versions (useful for legacy or longtime projects), but the process is complicated, limiting practical availability effectively to latest versions. | Fetches older documentation just-in-time, and always remains current, identical to latest docs. | | Search strategy | Uses either A) optionally token-limited RAG search to filter/preprocess docs which can be hit-or-miss, or B) dumps all content into LLM's context, resulting in excessive token-use. | Employs agentic discovery (as used by Claude Code itself) | | Developer Experience | Requires explicitly mention of context7 in every prompt. | Operates via defined rules/instructions, auto-selecting appropriate package name/version from lock file ("set and forget"). | | Support | Language/ecosystem independent. | Currently language/ecosystem-specific; planned additional languages/ecosystems and language-independent fallback mechanism soon. |