Muser has 5 tools with clear verb-noun naming (search_*, list_*, index_*) and reasonable descriptions (80-120 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the visible code, the server returns structuredContent with GalleryHit objects, but the LLM-facing schema is not declared. Parameter descriptions are adequate but lack constraints (e.g., no enum for 'mode' in search_by_color, no format spec for hex color). Error handling is minimal, no recovery guidance or actionable error messages visible. The server proxies a Python service; failure modes (service down, invalid image path, indexing errors) are not explicitly handled in the tool definitions.

• Document output schemas for all search tools. Specify that results include: path (string), score (float), dupes (array of strings), thumb (base64 data URI), full (base64 data URI). This enables LLMs to extract and chain results.
• Add enum constraint to search_by_color 'mode' parameter. List valid modes in the schema (e.g., 'exact', 'similar', 'palette') and in the description.
• Add format hint to search_by_color 'hex' parameter: 'Hex color code in #RRGGBB format (e.g., #FF5733)'. Include validation error message in tool implementation.
• Add error recovery guidance to all tools. E.g., 'If indexing fails, check that the folder path is valid and contains supported image formats (JPEG, PNG, WebP). Retry with a smaller folder.'
• For index_folder, add an optional 'dry_run' boolean parameter (default false) to preview what would be indexed without modifying the database. Or document idempotency: 'Safe to call multiple times; re-indexing updates embeddings for changed files.'

Inferred effective spec: 2025-06-18+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained