Cowork History
Search past Cowork sessions with keyword, macOS Spotlight, and semantic vectors so agents can reuse decisions and snippets from earlier work.
Overview
Cowork History MCP is a MCP server for the Build phase that hybrid-searches Cowork session history with FTS, Spotlight, and semantic vectors.
What is this MCP server?
- Hybrid retrieval: full-text search, Spotlight, and semantic vectors in one MCP server (v4.0.3)
- Local Ollama embeddings via OLLAMA_URL and EMBEDDING_MODEL (default nomic-embed-text)
- stdio MCP through PyPI cowork-history or matching MCPB release
- Targets Cowork session history rather than generic git or filesystem indexes
- Pairs lexical and semantic ranking for long-running solo builder threads
Community signal: 1 GitHub stars.
What problem does it solve?
You lose hours re-finding what you already decided in old Cowork threads because plain chat scroll and generic search do not surface the right passage.
Who is it for?
Mac-using solo builders on Cowork who run local Ollama and want agent-callable recall across long session archives.
Skip if: Linux-only shops without Cowork history, teams needing centralized cloud knowledge bases, or builders who refuse local embedding setup.
What do I get? / Deliverables
After install, your agent can query prior Cowork work through one MCP surface that blends keyword, Spotlight, and embedding hits.
- MCP search tools spanning FTS, Spotlight, and embedding-ranked hits
- Configurable local embedding model via EMBEDDING_MODEL
- stdio transport packages at version 4.0.3
Recommended MCP Servers
Journey fit
Primary shelf is agent-tooling because it extends what your coding agent can recall while you build; it also supports operate iterate and grow content when mining old threads. Hybrid FTS plus Spotlight plus Ollama vectors turns scattered session history into queryable MCP tools for the agent.
How it compares
Cowork-specific hybrid history MCP, not a general codebase RAG indexer or team Notion connector.
Common Questions / FAQ
Who is Cowork History MCP for?
Solo builders using Cowork with AI agents who need fast recall across many past sessions on their own machine.
When should I use Cowork History MCP?
Use it while building or iterating when the agent should cite prior Cowork decisions, debugging notes, or draft copy you already produced.
How do I add Cowork History MCP to my agent?
Install cowork-history from PyPI or the v4.0.3 MCPB bundle, point OLLAMA_URL at your Ollama instance, set EMBEDDING_MODEL if needed, and add the stdio server to your MCP client config.