# BetaBots-LLC/callimachus [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/BetaBots-LLC/callimachus  
**GitHub Stars:** 39  
**npm Downloads (last month):** 531  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/betabots-llc-callimachus

## Description
Local index and hybrid search (SQLite FTS5 + on-device vector KNN) over your AI coding-agent conversation history across 11 tools (Claude Code, Codex, Cursor, and more). Exposes searchthreads, searchcurrentproject, recentthreads, and getthread so any agent can recall its own past work.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "callimachus": {
    "command": "npx",
    "args": ["-y","callimachus-mcp"]
  }
}
```

## Documentation

## What BetaBots-LLC/callimachus MCP server does

BetaBots-LLC/callimachus MCP server gives coding agents access to a local catalogue of conversation history. Callimachus indexes threads from 11 supported coding-agent tools: Claude Code, Codex, Cursor, Gemini CLI, Qwen Code, Goose, OpenCode, Continue, Cline, Roo Code, and Kilo Code. The resulting data is stored in SQLite and can be queried without sending the indexed history to a hosted search service.

The main use is historical recall. An agent can look for prior work across all indexed threads, narrow results to the current project, inspect recent threads, or retrieve a selected thread. Search can also use file mentions, allowing a path such as `file:embed/mod.rs` to find conversations that discussed a particular file. Filters include source, project, subagents, starred threads, and tags.

## How it works

Keyword retrieval uses SQLite FTS5 with BM25 ranking. Semantic retrieval uses embeddings generated on the device with `fastembed`, then performs vector KNN queries through `sqlite-vec`. Callimachus combines the keyword and semantic result lists using Reciprocal Rank Fusion. Semantic search is not available immediately after installation: the conversation index must be populated first, followed by a separate semantic-index build.

The MCP sidecar and the desktop application use the same local `index.db` and share the desktop core library. A background file watcher keeps the index current after sources have been configured. The server can also write selected information back to Callimachus memory, including completed TODOs, decisions, gotchas, and session snapshots, rather than modifying project files directly.

## Setup and configuration

The project publishes signed desktop builds for macOS, Windows, and Linux through GitHub Releases. It can also be built from the repository with pnpm and the Tauri development workflow. The repository is a pnpm and Turborepo monorepo; development setup includes `pnpm install` followed by `pnpm desktop:dev`.

After the first launch, configure the conversation sources in Settings or trigger Reindex. Build the semantic index separately when vector search is needed. Reindexing and semantic-index construction run as distinct background jobs and do not hold the database write lock at the same time.

No cloud search credential is required for the local index or MCP history queries. Optional LLM features can use local Ollama, an existing Claude Code or Codex CLI subscription, or a configured cloud provider. Stored provider credentials use the operating system credential store.

## Tools and capabilities

The documented MCP-facing capabilities include:

- `searchthreads` for searching indexed conversations.
- `searchcurrentproject` for restricting retrieval to the active project.
- `recentthreads` for finding recently indexed conversations.
- `getthread` for reading a selected conversation.
- `linked_commits` for retrieving inferred relationships between threads and Git commits.
- `snapshot_session`, `list_snapshots`, and `load_snapshot` for creating and resuming session checkpoints.
- `check_decision` for surfacing an existing decision before an agent revisits a proposal.

The exact availability of memory-writing and session features depends on the bundled application version and its local data. The server is intended for agents that can call MCP tools during a session; the README does not provide a client-specific configuration command.

## Limitations and notes

Results depend on which sources have been configured and indexed. A new installation starts with an empty index, and semantic retrieval requires the additional semantic-index build. The semantic model and vector search run locally, but optional distillation, RAG answers, and chat can send prompts to the LLM provider selected by the user.

This project is licensed under AGPL-3.0. The provided material does not identify a standalone package-install command for `callimachus-mcp`; it describes the MCP server as a bundled sidecar of the desktop application.

_Full upstream README: https://allmcps.com/mcp/betabots-llc-callimachus/readme_