SafeSkill 92/100

docent

Semantic + BM25 Document search — an experimental MCP server written in Rust that indexes markdown documents, letting agents query why code looks the way it does.

  files ──▼── index (in-memory) ──▶  MCP server  ◀──── query
                              (HTTP)

Quick Start

docent init              # generate docent.toml config
docent serve             # start MCP server on port 7878 (indexes doc_dirs in the background)

Open http://localhost:7878 for the built-in Web UI.

Usage

Command	Description
docent init	Generate a docent.toml config file
docent serve	Start the MCP server (streamable HTTP); indexes in the background
docent list-models	List supported embedding models

Flags: --config <path> (default ./docent.toml).

How It Works

1. Sources — Reads markdown files from [index] doc_dirs (default: ./).
2. Section-aware chunking — Splits documents into chunks, preserving heading structure.
3. Embedding — Converts chunks to vectors via fastembed (configurable model).
4. In-memory index — Builds an in-memory semantic + BM25 index on serve startup; nothing is persisted to disk.
5. Auto-refresh — Watches doc_dirs for changes (debounced) and incrementally reindexes edited files. Stale results during reindex are flagged via SearchResult.stale = true.
6. Semantic + BM25 search — Hybrid scoring with configurable algorithm.
7. MCP server — Exposes search_doc tool over streamable HTTP.

Install

TBC

Documentation

• Development Guide