snapshot file gets overwritten

[berekvolgyipeter]

Describe the bug
There is a ~/.context/mcp-codebase-snapshot.json file which stores the indexing statuses of codebases. When indexing a new codebase from a different claude-code project, the file gets overwritten rather than appending the new codebase it, making claude-context forget that the codebase was indexed. This is particularly problematic if you use a self hosted milvus instance, because in this case there is no syncing with the cloud.

I also observed this issue when trying to index multiple codebases concurrently. The snapshot file gets overwritten every time and at the end of the concurrent indexing, there's only one codebase snapshot at the json: the codebase that's indexing finished last.

Suggestions:
When writing into ~/.context/mcp-codebase-snapshot.json file, check if it has contents and merge it to the in-memory content before writing.
Also lock the file while it's being written to protect it from overwriting by concurrent claude-context processes.
Or maybe do sync with self hosted milvus instances as well.

What's your MCP Client Setting

"claude-context": {
    "type": "stdio",
    "command": "sh",
    "args": [
        "-c",
        "export $(grep '^OPENAI_API_KEY=' .env | xargs) && exec npx @zilliz/claude-context-mcp@latest"
    ],
    "env": {
        "OPENAI_API_KEY": "${OPENAI_API_KEY}",
        "MILVUS_ADDRESS": "localhost:19531"
    }
}

Thank you if you address this.

[zc277584121]

Thanks for reporting this issue!

This has been fixed in v0.1.5 (PR #282). The saveCodebaseSnapshot() method now uses a read-merge-write approach with file locking, so concurrent MCP server processes will no longer overwrite each other's snapshot entries.

Additionally, we fixed a related issue where a codebase stuck in "indexing" state (e.g., due to process crash) would permanently block re-indexing. Now:

• force=true can bypass and clear stale indexing state
• On MCP server restart, interrupted "indexing" entries are automatically reset to "indexfailed"

Please update to v0.1.5 and test. If no further issues are reported, this will be closed after a period of time.