Code Memory
Index your repo locally for semantic code search and Git-aware recall so agents find the right file without sending the whole tree.
Overview
io.github.kapillamba4/code-memory is an MCP server for the Build phase that provides offline local semantic code search with Git history for agents.
What is this MCP server?
- Local semantic code search with Git history context (PyPI code-memory v1.0.33)
- Works offline with no API key required for core search workflows
- Configurable HuggingFace embedding model via EMBEDDING_MODEL env var
- Logging verbosity via CODE_MEMORY_LOG_LEVEL (DEBUG through ERROR)
- stdio MCP transport suited to privacy-sensitive or air-gapped dev machines
- Version 1.0.33 on PyPI identifier code-memory
- stdio transport
- Documented env vars CODE_MEMORY_LOG_LEVEL and EMBEDDING_MODEL
Community signal: 41 GitHub stars.
What problem does it solve?
Agents waste turns grepping and guessing because they cannot semantically search your private repo without huge context dumps.
Who is it for?
Privacy-conscious indie devs with medium-to-large repos who want MCP-powered semantic navigation in Claude Code or Cursor.
Skip if: Teams that want a hosted, multi-repo enterprise search product with centralized admin and SSO.
What do I get? / Deliverables
Your agent queries a local code-memory index for relevant snippets and history-backed context while you keep code on your machine.
- Running stdio MCP code-memory server against your repo
- Semantic retrieval results tied to local code and Git context
- Tuned logging and embedding configuration for your machine
Recommended MCP Servers
Journey fit
Semantic navigation accelerates implementation when the codebase is large enough that filenames and grep stop scaling. agent-tooling is the shelf for MCP servers that extend the coding agent’s memory and retrieval inside Build.
How it compares
Local semantic code MCP index, not a cloud documentation crawler or generic web RAG stack.
Common Questions / FAQ
Who is io.github.kapillamba4/code-memory for?
Solo builders and small teams who want offline, API-key-free semantic code search for MCP-connected coding agents.
When should I use io.github.kapillamba4/code-memory?
Use it during Build when navigating unfamiliar modules, tracing Git-related changes, or reducing blind edits in a large codebase.
How do I add io.github.kapillamba4/code-memory to my agent?
Install the code-memory package from PyPI, point your MCP client at the stdio server entry, and set optional EMBEDDING_MODEL and CODE_MEMORY_LOG_LEVEL as needed.