Claude Faf MCP

What Wolfe-Jam/claude-faf-mcp MCP server does

Wolfe-Jam/claude-faf-mcp MCP server manages persistent project context in the .faf format, described in the repository as Foundational AI-context Format. The context is intended to capture both repository information and human-supplied intent, including the six project questions commonly represented as who, what, why, where, when, and how.

The server exposes a core set of MCP tools for creating, reading, checking, scoring, and synchronizing this context. It can initialize project DNA, inspect a project’s technology stack, derive information from a README, or load context from a GitHub repository URL. Human context can be added separately rather than inferred from source files.

Its scoring tools report an AI-readiness result from 0 to 100% with a breakdown. faf_bench compares agent grounding with and without the .faf file and emits a receipt describing the result. The available tools also include project status, environment inspection, file reads and writes, and an explanation of the .faf format.

How it works

The repository treats .faf as the primary context source. faf_init, faf_auto, and faf_quick provide different entry points for creating project context, while faf_go guides the user through an interview intended to improve the result. faf_readme and faf_git add evidence from existing project documentation or a remote GitHub repository.

After context is created, faf_check validates its structure and faf_doctor diagnoses and fixes common problems. faf_score measures the resulting context, and faf_status provides a project-level view. The README also describes a /faf prompt for Claude Desktop that checks, scores, and synchronizes a project.

Wolfe-Jam/claude-faf-mcp MCP server can synchronize .faf into CLAUDE.md. Its synchronization options also cover AGENTS.md, .cursorrules, GEMINI.md, and .github/copilot-instructions.md. A tri-sync operation connects .faf, CLAUDE.md, and MEMORY.md.

Setup and configuration

The project is published as the npm package claude-faf-mcp and is licensed under MIT. The repository describes the server as a TypeScript MCP server and includes an npm package badge. A package-runner configuration can start it with:

npx -y claude-faf-mcp

The README distinguishes between a core tool set and an expanded set: it states that the server has 12 core tools and 34 tools when FAF_TOOLS=all is used. The supplied material does not document other required environment variables, credentials, API keys, or hosted service endpoints.

Tools and capabilities

The listed capabilities include:

• Initialize, auto-populate, or quickly create project DNA.
• Extract context from README files and GitHub repository URLs.
• Add human context using the six project questions.
• Score AI readiness and benchmark grounding with a receipt.
• Validate .faf structure and diagnose common issues.
• Read and write files, inspect the environment, and show project status.
• Import or export AGENTS.md, .cursorrules, GEMINI.md, and Conductor content.
• Synchronize project context into files used by several coding assistants.

Wolfe-Jam/claude-faf-mcp MCP server is most relevant when repository code alone does not express project intent, or when several agent clients need a shared, generated context surface.

Limitations and notes

The provided material does not specify operating-system requirements, supported Node.js versions, or a complete installation walkthrough. Although the tools include GitHub URL extraction, no GitHub credential requirement is stated. The tri-sync capability is identified as a Pro feature in the tool list, while also being described there as free for developers; the supplied information does not define any separate paid plan or feature limits. The README refers to version-specific composition with faf-cli, but the exact dependency installation and version-management procedure is not included here.