Faf MCP

What Wolfe-Jam/faf-mcp MCP server does

Wolfe-Jam/faf-mcp MCP server uses project.faf as a central project-context file for AI-assisted development. The server can initialize, read, validate, update, score, and synchronize this context. Its purpose is to keep one project definition aligned with platform-specific files used by coding agents and IDEs.

Supported context targets include AGENTS.md, .cursorrules, GEMINI.md, and CLAUDE.md. Interoperability tools also cover Conductor directory structures and creating a FAF file from a GitHub repository URL. The repository identifies the FAF format with the media type application/vnd.faf+yaml.

How it works

The server exposes MCP tools that operate on local project context and related platform formats. A typical workflow starts with faf_init, uses faf_read or faf_score to inspect the project, and then applies faf_sync or faf_bi_sync to propagate changes. The bi-directional workflow keeps the .faf file and supported agent instruction files aligned rather than treating each file as an independent source of truth.

The default configuration presents 15 essential tools. Setting FAF_TOOLS=all exposes the full set of 29 tools, while tools remain callable by name. The README also documents a CLI fallback through faf-cli for operations such as extracting project information from a README or merging YAML content.

Setup and configuration

For a local stdio deployment, run:

bunx faf-mcp

An MCP client can register it with the bunx command and faf-mcp argument. The README lists configuration locations for Cursor, Windsurf, Cline, and VS Code. Cursor also has a direct installation link. Claude Desktop users are directed to the separate claude-faf-mcp project rather than the standard IDE configuration shown for this repository.

A hosted Streamable HTTP endpoint is documented at https://ide.faf.one/mcp/v1. The project also documents a Vercel deployment option, with /health, /info, and /mcp/v1 routes available on a deployed instance.

Tools and capabilities

Documented tool groups include:

• Core project operations: faf_init, faf_score, faf_sync, faf_bi_sync, faf_read, and faf_write.
• Platform interoperability: faf_agents, faf_cursor, faf_gemini, faf_conductor, and faf_git.
• Cloud context operations: faf_cloud_publish, faf_cloud_fetch, faf_cloud_list, faf_cloud_search, and faf_cloud_share.

Cloud tools can publish, retrieve, list, search, and share named context records through mcpaas.live. The README describes shareable URLs and fetching context back into a local project.faf file.

Limitations and notes

Wolfe-Jam/faf-mcp MCP server is focused on project context management, not general-purpose code execution or repository hosting. The README describes multiple deployment paths, but it does not specify authentication requirements, hosted-service pricing, or access controls for the hosted endpoint. The repository documentation also distinguishes the default 15-tool surface from the 29-tool full set, so available tools depend on FAF_TOOLS configuration.

The documented synchronization targets are specific formats and platforms; unsupported agent formats may require the CLI fallback or another integration. The project is MIT licensed.