M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
C

Claude FAF

Community
Persistent project context for Claude. IANA-registered .faf format.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #272 of 438
Stars
★ 23 Popular
Transport
stdio (local process)
Runtime
Node.js · Bun
Credentials
No credential needed
License
MIT
Last commit
Tools
27
49FMRS · D

This server provides persistent context and memory capabilities with high integration, but users should be cautious about file operations; overall valuable.

Strongest · Setup experience 12/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
8/20
Maintenance
11/20
Documentation
10/20
Setup experience
12/20
Why each score
Reliability 8/20
README and manifest are untrusted evidence and are internally inconsistent: the repo description says '32 tools + 1 prompt', the README header says 13 Core/35 total, body text says 'Core of 12' and '36 total'; the manifest version 5.22.1 conflicts with README mcpb links (5.13.0). No verifiable CI workflow files or committed test code are supplied in the source material (only badge claims of '572 tests'), so under static calibration the score cannot exceed 12. Claims like '19ms average execution' and 'deterministic scoring' are statically unverifiable.
Security and permissions 8/20
No red-line findings: no malware, credential theft, or covert exfiltration; README claims fully local operation with no telemetry, the mcpb includes a sha256, and v5.8.0 notes claim path-confined file access. However, faf_read/faf_write are described as 'Read any file / Write any file' — an overly broad write surface with no demonstrated mandatory confirmation for dangerous operations; faf_setup modifies user Claude Code settings (hooks), and the 'preview first' safeguard cannot be statically verified. Deducted for unverifiable scoping and confirmation mechanisms.
Maintenance 11/20
Repo is not archived, MIT license present, rapid release cadence (5.8→5.22+), multi-channel distribution (npm/mcpb/Smithery), manifest includes repository id, and there is at least one tracked issue. But the very fast iteration with version drift between docs and manifest, absence of evidence for a SECURITY.md or security-response channel, and a supply-chain dependency on faf-cli ^7.7.0 composition limit the score; governance evidence is thin.
Documentation 10/20
README is extensive: install paths, tool tables, scoring tiers, privacy statement, citations, ecosystem list, plus faf.one site and a linked tools reference. But it has real defects: contradictory tool counts (12/13/32/35/36), manifest-vs-README version mismatch, ambiguous 'Pro feature, free for developers' semantics, heavy unverifiable marketing claims (DOIs, IANA registrations not verifiable from supplied material), and no troubleshooting or known-limitations section.
Setup experience 12/20
Multiple install paths are provided: one-click .mcpb, bunx/npx config snippet for claude_desktop_config., global npm install, and a hosted Smithery URL — the main flow is clear. However, the README's own mcpb link points to 5.13.0 instead of manifest version 5.22.1, the bunx path assumes bun is installed (not universal), and the hosted endpoint adds an external dependency; consistent with the static setup cap of 15.

Static review · not runListed 2026-09-08

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or code

Best for

  • Developers using Claude Desktop
  • Teams needing persistent, versioned project context
  • Those requiring cross-session AI memory

Not for

  • Personal projects without AI context needs
  • Users not using AI tools that support .faf format
  • Users who may be overwhelmed by extended tools

Required permissions

  • Read and write project files
  • Runs locally, no network egress

Risks and side effects

  • File overwrites; review each sync
  • Some tools deprecated, but core stable

Setup

Before you start

Runtime：Node.js · Bun

Download the .mcpb file and double-click to install, or use npm install -g claude-faf-mcp, then configure claude_desktop_config.json with: {"mcpServers":{"faf":{"command":"bunx","args":["claude-faf-mcp"]}}} and restart Claude Desktop.

claude_desktop_config.json 
{"mcpServers":{"faf":{"command":"bunx","args":["claude-faf-mcp"]}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "faf": {
      "command": "bunx",
      "args": [
        "claude-faf-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add faf -- bunx claude-faf-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After install, faf_* tools such as faf_init and faf_score should appear in your client's tool list; you can also type the /faf prompt and the server should check and score your project, confirming the connection works.

Troubleshooting

  1. Ensure npm package installed globally, check command in claude_desktop_config.json, ensure network access to npm or GitHub, or check Claude logs.

Things to try

Once connected, you can ask your AI assistant things like:

  • Initialize project.faf for this repo with faf_init
  • Run faf_auto to auto-detect my stack and populate the project context
  • Score this project — what's my AI-readiness percentage?
  • Run faf_sync to sync .faf to CLAUDE.md

Tools 27

faf_init writes
Initialize project DNA.
faf_auto writes
Auto-detect stack and populate context.
faf_go writes
Guided interview to Gold Code.
faf_bench read-only
Benchmark AI grounding cold vs .faf, with a ✪ receipt.
faf_score read-only
AI-readiness score (0-100%) with breakdown.
faf_doctor writes
Diagnose and fix common issues.
faf_sync writes
Sync .faf to CLAUDE.md or AGENTS.md etc.
faf_context read-only
Get or display current context.
Show 19 more tools
faf_trust read-only
Verify score determinism.
faf_about read-only
What is .faf?
faf_etch writes
Extract or embed context.
faf_recall read-only
Recall memory.
faf_quick writes
faf_readme read-only
faf_formats read-only
faf_git read-only
faf_human_add writes
faf_check read-only
faf_tri_sync writes
faf_agents writes
faf_cursor writes
faf_gemini writes
faf_conductor writes
faf_read read-only
faf_write writes
faf_status read-only
faf_debug read-only

Use cases

Provide consistent project context in every Claude session.
Get new team members up to speed immediately after cloning the repo.
Persist project facts across sessions for AI memory.
Ensure AI follows project-specific conventions.

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

Claude FAF is an MCP server that provides persistent project context and memory for Claude. It utilizes the IANA-registered .faf format (application/vnd.faf+yaml) to store a project's DNA, including stack, goals, and 6Ws. It offers tools to initialize, auto-detect, score, sync, and audit context, ensuring AI starts each session with a coherent understanding. It provides 13 core tools by default, extensible to 35. Interoperability with multiple AI environments is achieved through writing files like CLAUDE.md, AGENTS.md, etc.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 4960584d9291 Data synced 2026-10-19 Read the FMRS scoring method