fix(mcp): refuse second writer for same palace

[fatkobra]

What does this PR do?

Closes #1818

• Add a process-lifetime per-palace MCP writer guard.
• Reuse the existing mine palace lock so an MCP process does not become a second writer against the same Chroma palace.
• Let peer MCP servers continue serving read tools.
• Refuse mutating tools with a structured JSON-RPC error when another writer is active.
• Add regression tests for write gating and read-tool pass-through.

Why

Two long-lived MCP stdio servers can point at the same Chroma palace. Per-operation write locks serialize individual operations, but they do not invalidate another process's already-open Chroma PersistentClient and in-memory HNSW/FTS state. Holding the per-palace writer lease for the MCP server lifetime prevents a second MCP writer from silently corrupting the same palace.

How to test

• python3 -m py_compile mempalace/mcp_server.py tests/test_mcp_server.py
• python3 -m ruff format --check mempalace/mcp_server.py tests/test_mcp_server.py
• python3 -m pytest tests/test_mcp_server.py
• python -m pytest tests/ -v

Checklist

• Tests pass (python -m pytest tests/ -v)
• No hardcoded paths
• Linter passes (ruff check .)

[gemini-code-assist]

Code Review

This pull request introduces an MCP peer-writer guard to prevent multiple long-lived Chroma clients from using stale in-memory state by enforcing a per-palace writer lease. Mutating tools are restricted to read-only mode if the writer lock cannot be acquired, and corresponding unit tests have been added. Feedback on these changes highlights that an unexpected exception during lock acquisition is not cached, which would cause subsequent mutating tool calls to repeatedly attempt to acquire the lock and flood the logs; introducing a state variable to cache this failure is recommended.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[fatkobra]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces an MCP peer-writer guard to prevent concurrent mutating operations across multiple processes, protecting against stale in-memory Chroma/FTS state. It defines a set of mutating tools, implements a lock acquisition mechanism with caching for failures, and refuses mutating requests when the server is in read-only mode. Additionally, corresponding unit tests have been added to verify this behavior. I have no feedback to provide as there are no review comments to address.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[fatkobra]

/gemini review

[fatkobra]

Adds startup SQLite/FTS5 integrity gate (#1818, gap flagged by @99LevelsUp)

The writer guard above stops new concurrent corruption. This closes the remaining gap: a palace that's already corrupt when the server starts (e.g. from a prior killed writer) was served silently with no integrity check on open.

Read-only integrity probe (repair.sqlite_integrity_errors()) runs once on startup, cached in module state.
Mutating tools refuse with a dedicated error code (-32002) and structured diagnostic when integrity is bad; mempalace_status and mempalace_reconnect stay available.
mempalace_reconnect re-runs the check, so an out-of-band mempalace repair --yes clears the gate without a server restart.
No auto-heal — corruption is surfaced and the operator repairs explicitly, rather than the server mutating the database on its own.

[gemini-code-assist]

Code Review

This pull request introduces an SQLite integrity gate and a peer-writer guard to the MemPalace MCP server to prevent concurrent writes and handle database corruption gracefully. It adds several helper functions for checking integrity and acquiring locks, updates existing tools like tool_status and tool_reconnect to utilize these guards, and includes comprehensive unit tests. The feedback highlights potential TypeError crashes if _config.palace_path is None when constructing file paths or performing integrity checks, and suggests registering an atexit handler to ensure the acquired palace lock is properly released on process termination.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[fatkobra]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces an SQLite integrity gate and a peer-writer guard to the MemPalace MCP server. The integrity gate runs a quick check on startup to prevent serving malformed indexes, while the peer-writer guard holds a per-palace lock to prevent concurrent mutating operations. Comprehensive unit tests have been added to verify these behaviors, including handling edge cases like unconfigured palace paths. I have no further feedback to provide as no review comments were submitted.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.