Performance: MCP Search Memory tool responds slowly when running Core locally (docker-compose)

Summary

The MCP Search Memory tool responds slowly. This is observed while using the MCP server in Claude Code against a local instance of RedPlanetHQ/core (via docker-compose). Investigate and, if possible, improve performance.

Context

• Tool: MCP Search Memory
• Runtime: Local core via docker-compose
• Observation: Responses take longer than expected.

Impact

• Slower development loop when using memory search.
• Latency may indicate a performance issue in the MCP server, Core, or the integration path between them.

Environment

• Core: local via docker-compose
• Client: Claude Code

Steps to Reproduce

1. Start core locally using docker-compose.
2. Use Claude Code connected to the local MCP server.
3. Run the MCP Search Memory tool.
4. Observe response latency.

Expected Behavior

Search results return promptly with low latency during normal local development.

Actual Behavior

Responses are noticeably slow.

Notes / Hypotheses (to be confirmed)

• Local Core setup (container resource limits, I/O, or networking) could contribute.
• Overhead in request/response serialization between Claude Code and Core.

What to Investigate

• Measure end-to-end latency (client → MCP → Core → MCP → client).
• Capture timings per hop (MCP server, Core API, storage/search layer).
• Check Docker CPU/RAM limits and container stats during search.
• Review logs for slow queries, retries, or timeouts.
• Compare performance against a non-docker local run (if applicable).
• Profile MCP server handling for search requests.
• Verify index/state warmup behavior and caching.

Potential Mitigations (pending findings)

• Tune Docker resources (CPU/RAM/IO).
• Add/request caching or index warmup.
• Optimize query path or payload size.
• Parallelize or batch internal calls if applicable.

Requested Outcome

• Root cause identified.
• Concrete fix or mitigations with before/after latency numbers.
• If external (Claude Code MCP) issue, document and track upstream.

Start by reproducing the MCP Search Memory request through Claude Code against Core via docker-compose, then measure latency across the client, MCP server, Core API, and storage/search layer. Review container stats and logs while profiling the request path; done means identifying the root cause and documenting a concrete fix or mitigation with before-and-after latency numbers.