Quickstart
The public package is named repo-memory-mcp-ai. The server executable remains repo-memory-mcp.
pip install repo-memory-mcp-ai repo-memory doctor repo-memory-mcp
For a zero-install run, use:
uvx --from repo-memory-mcp-ai repo-memory-mcp
MCP configuration
Add this stdio server to the configuration file used by your MCP host:
{
"mcpServers": {
"repo-memory": {
"command": "uvx",
"args": ["--from", "repo-memory-mcp-ai", "repo-memory-mcp"]
}
}
}Agent workflow
- At session start, call
get_startup_context(). - When a durable correction is made, call
remember_correction(topic, lesson). - When a project convention is established, call
remember_rule(rule). - When architecture changes, call
remember_decision(decision, rationale). - Search before repeating work with
recall_memory(query). - Archive obsolete guidance with
forget_memory(memory_id).
Core tools
| Tool | Purpose |
|---|---|
get_startup_context | Return concise, high-value repository guidance. |
remember_correction | Persist a lesson from a correction or bug fix. |
remember_rule | Persist a convention, constraint, or project rule. |
remember_decision | Persist an architecture or implementation decision. |
recall_memory | Search active memory with deterministic local relevance scoring. |
update_memory | Correct existing guidance without replacement records. |
forget_memory | Archive obsolete memory for audit-friendly retention. |
export_memory / import_memory | Move validated memory between safe copies of the same repository. |
Privacy and storage
Memory is stored locally at .repo-memory/memory.db in the detected repository. The server does not require API keys, cloud accounts, external embeddings, remote databases, or telemetry. Do not store credentials, tokens, passwords, or private keys.
Troubleshooting
My client cannot start the server
Run uvx --from repo-memory-mcp-ai repo-memory-mcp from a terminal. If it works there, check that your client uses the exact stdio JSON configuration and has access to uvx on its PATH.
Memory is not appearing
Confirm the client called a remember tool and that you are working inside the same repository. Run repo-memory doctor and inspect .repo-memory/.