MCP

Agents can search the journal without managing it. pnpm mcp is a stdio server. Search and tags use data/journal.sqlite (FTS5 + sqlite-vec) when that file has rows; otherwise it walks sidecars. No journal-content writes; opening the derived index can migrate its cache schema. The API does not need to be running. Rebuild with pnpm journal:index.

Tools

Tool	What it does
`dictawhisper_search`	Search notes
`dictawhisper_get_note`	Fetch one sidecar
`dictawhisper_list_tags`	List tags
`dictawhisper_recent`	Recent notes

This repo ships .cursor/mcp.json. Elsewhere:

{
  "mcpServers": {
    "dictawhisper": {
      "command": "node",
      "args": ["--experimental-strip-types", "src/mcp.ts"],
      "cwd": "/absolute/path/to/dictawhisper"
    }
  }
}

Current compatibility

Existing search/get/recent/tag tools remain available. Sidecars may now include entryId, displayTitle, recordedDate and date provenance, MayDos and their status/extraction metadata. Legacy sidecars remain readable. MCP uses its existing capped read-only tool contracts; HTTP page=1 provides UI pagination and is not an implicit change to MCP inputs.

Settings → Integrations supplies the executable path, source or packaged MCP entry point, and loaded DICTA_CONFIG for this installation. ExoMetaCortex synchronization is planned; no export or automatic sync is connected yet.