rapidbm25
Local BM25 keyword search over texts or a code and docs folder, with a Rust core
- Platforms
- CLI, Linux, macOS
- Runs
- Locally
- Version
- 0.2.0
- Listing updated
- 4 days ago
Similar MCP servers
About rapidbm25
rapidbm25 is an MCP server for developer tools. It runs locally on your machine and is started by your AI client.
Behind this MCP server is rapidbm25, a Rust-native stand-in for the rank_bm25 Python classes BM25Okapi, BM25L and BM25Plus. The README claims bit-for-bit identical scores and roughly 1,470 times faster get_scores, and the server wraps the same engine so an agent can run keyword ranking without a network call.
The server indexes a folder through bm25_index_path, which is backed by a live index that supports add, delete and update without rebuilding. Two environment variables control it: RAPIDBM25_ROOTS limits which directories may be read (default the working directory) and RAPIDBM25_CACHE sets the cache folder (default ~/.cache/rapidbm25).
The library side adds optional tokenizers with stemming and CJK bigrams, metadata filters, batch search, save and load, and reciprocal rank fusion for hybrid ranking. Prebuilt wheels cover CPython 3.9 to 3.15 on Linux, macOS and Windows.
Install
Run in a terminal
claude mcp add rapidbm25 -- uvx rapidbm25Add to claude_desktop_config.json
{
"mcpServers": {
"rapidbm25": {
"command": "uvx",
"args": [
"rapidbm25"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"rapidbm25": {
"command": "uvx",
"args": [
"rapidbm25"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"rapidbm25": {
"type": "stdio",
"command": "uvx",
"args": [
"rapidbm25"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"rapidbm25": {
"command": "uvx",
"args": [
"rapidbm25"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"rapidbm25": {
"command": "uvx",
"args": [
"rapidbm25"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Configuration
RAPIDBM25_ROOTSDirectories bm25_index_path may read (os.pathsep-separated). Default: the working directory.RAPIDBM25_CACHEFolder-index cache directory. Default: ~/.cache/rapidbm25.
Tools this server provides (1)
Taken from the project documentation; the running server may differ.
- bm25_index_path
- Index a folder of code or docs for BM25 keyword search
What it does
- Drop-in rank_bm25 scores
- Matches rank_bm25 scores exactly, with a compat option for the project's git-main behaviour.
- Live index
- Add, delete and update documents without rebuilding posting lists.
- Metadata filters
- Restrict searches with operators such as $eq, $in and $gt.
- Optional tokenizers
- Unicode words, stop lists, Snowball stemming and CJK bigrams.
- Rank fusion
- Combine several rankings with reciprocal rank fusion.
Good for
- Search a code or docs folder by keyword from an agent
- Replace rank_bm25 in a Python pipeline for speed
- Combine keyword and vector rankings with rank fusion
- Keep an incrementally updated search index
rapidbm25 FAQ
Does it need a network connection?
No; the registry entry describes it as local with no network.
Which directories can the server read?
Those listed in RAPIDBM25_ROOTS, defaulting to the working directory.
Is it identical to rank_bm25?
Scores match bit for bit; tie order and a few edge cases differ, as documented in the README.
How do I add rapidbm25 to Claude Code?
Run this command in your terminal: claude mcp add rapidbm25 -- uvx rapidbm25
What are the best alternatives to rapidbm25?
The closest alternatives to rapidbm25 are rag-mcp, WARM, MiniRAG, MarkIndex and Searchcraft.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.
