#1212th of 86 in MCP servers for developers
fff
Fast typo-tolerant file and content search for MCP clients, Pi and Neovim
- Stars
- 11.2k
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Sep 2026
- Language
- Rust
4,155 on 10 Apr 202611,192 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
fff is a Rust file search library with a resident in-memory index, frecency ranking, git-status annotations and a background watcher. It ships as an MCP server (tools grep, find_files, multi_grep) for Claude Code, Codex, OpenCode, Cursor and Cline, as a Pi agent extension, and as the fff.nvim picker. Search is typo-tolerant, with fuzzy fallback on zero matches and a definitions-only mode.
Who it is for: Users of coding agents or Neovim who want faster, ranked file and grep search
Strengths
- Index stays resident across tool calls, so repeated searches avoid re-scanning the repo
- Frecency ranking and git status tags surface files you are actively editing
- Definitions-only search with staged fallback from exact to fuzzy matches, with notices
- Prebuilt binaries via install script or Homebrew; same core serves MCP, Pi and Neovim
Weaknesses
- Benchmarks in the README show mixed results: one held-out model was slower and used more tokens
- Install is a curl-piped shell script or Homebrew; no container image
- Agents need a prompt hint in CLAUDE.md to prefer fff over built-in search
- Index is scoped to a git-indexed directory; globs cannot override ignore rules
What it needs
- no GPU
- Needs Rust (fallback cargo build for Neovim plugin), git
Quick answers
Is fff open source?
Yes. Its code is on GitHub under the MIT license.
Also in MCP servers for developers
See all 86| Rank | Project | Score |
|---|---|---|
| 1 | codegraphLocal code knowledge graph that coding agents query over MCP | 96 out of 100 |
| 2 | gitnexusCode knowledge graph exposed to coding agents through MCP tools | 96 out of 100 |
| 3 | serenaMCP server giving coding agents symbol-level code search, editing and refactoring | 91 out of 100 |
| #4 | n8n-mcpMCP server exposing n8n node docs, templates and workflow management to AI assistants | 90 out of 100 |
| #5 | context7MCP server and CLI that feed current library docs to coding agents | 89 out of 100 |