Codebase Memory
Persist a codebase knowledge graph across agent sessions so architecture and symbol relationships survive restarts and context compaction.
Overview
Codebase Memory is an MCP server for the Build phase that keeps a persistent codebase knowledge graph available to your agent across session restarts and context compaction.
What is this MCP server?
- Persistent codebase knowledge graph stored locally for the project
- Designed to survive session restarts and LLM context compaction
- stdio MCP transport via published mcpb binaries (v0.1.3)
- Release artifacts for darwin arm64/amd64 and linux amd64
- GitHub releases with checksum-published packages
- Version 0.1.3 with mcpb registry packages
- Published binaries for 3 platform targets (darwin arm64, darwin amd64, linux amd64)
- Transport type stdio per server manifest
Community signal: 3.1k GitHub stars.
What problem does it solve?
Coding agents lose repository structure and prior exploration whenever sessions reset or context is trimmed, forcing expensive re-reads of the same files.
Who is it for?
Solo builders on non-trivial repos who live in Claude Code or Cursor and want local, graph-shaped codebase memory without sending the whole tree to a cloud indexer every time.
Skip if: Greenfield hello-world repos, teams that forbid local graph databases, or builders who only need ephemeral chat without structural persistence.
What do I get? / Deliverables
After you install the mcpb binary and enable stdio MCP, your agent can reuse a durable codebase knowledge graph instead of rebuilding understanding from scratch each session.
- Running stdio MCP server maintaining a codebase knowledge graph on disk
- Improved cross-session recall of repository structure for the agent
Recommended MCP Servers
Journey fit
Long Build cycles on real repos need memory that outlives a single chat; this MCP targets durable codebase understanding while you integrate agent tooling. Agent-tooling covers MCP servers that extend what the agent remembers about your tree; a knowledge graph is structural memory, not a one-off lint fix.
How it compares
Local codebase knowledge-graph MCP via stdio binaries, not a hosted retailer API or document-only RAG SaaS.
Common Questions / FAQ
Who is Codebase Memory for?
Individual developers and indie teams using MCP-enabled agents who need persistent structural memory of a codebase across long build and iterate cycles.
When should I use Codebase Memory?
Use it during build and ongoing iteration when session restarts or context compaction keep wiping your agent’s map of modules, dependencies, and prior code exploration.
How do I add Codebase Memory to my agent?
Download the matching v0.1.3 mcpb release for your OS from GitHub, configure stdio MCP in your client to launch the binary, and point it at your project workspace.