codebase-memory-mcp

Use it when

• Your coding agent burns huge token budgets grepping file by file through a large repository to answer structural questions like what calls this function
• You need persistent, queryable architecture knowledge (call graphs, routes, impact analysis, dead code) that survives across agent sessions and can be shared with teammates as a committed graph artifact

What it solves

Not the fit when

• Editing or refactoring code
• Runtime profiling or tracing
• Cross-session conversational memory
• Natural-language answering without an MCP-capable agent
• Running the remote shell installer without reviewing its account-wide configuration, instruction, skill, and hook changes
• Committing or sharing generated graph artifacts without checking whether they disclose proprietary identifiers, routes, relationships, or architecture
• Assuming repository benchmark speed, accuracy, or token reductions for a different codebase or language mix
• editing or refactoring code
• runtime profiling or tracing
• cross-session conversational memory
• natural-language answering without an MCP client LLM

Install

Prefer downloading a release archive, verifying it against `checksums.txt`, extracting it, and reviewing/running the included installer. The one-line `curl ... | bash` path is documented but executes mutable remote code and also writes agent configuration, instructions, skills, and hooks.

Invoke

Restart your coding agent and say 'Index this project'; the agent then calls MCP tools such as get_architecture, trace_path, search_graph, semantic_query, detect_changes, or Cypher-like queries. CLI mode: codebase-memory-mcp cli search_graph '{"project": "my-project", "name_pattern": ".*Handler.*"}'

Alternatives

No reviewed alternatives recorded yet.