MCP server for claude-brain: exposes the knowledge store to Claude Code
claude-brain has well-structured tool definitions with consistent naming (verb_noun pattern), complete input schemas using schemars, and clear parameter descriptions. All 10 tools follow the verb-first convention (stats, store_entry, update_entry, delete_entry, search_entries, get_project_context, get_entry, list_entries, list_technologies, list_tags). Schemas are properly typed with serde/schemars. However, tool descriptions are brief (avg ~80 chars), lacking context on WHEN to use each tool or dependencies between them. Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent from visible code. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Parameter descriptions are adequate but lack format constraints (e.g., limit ranges, enum values for type/technology filters).
Delete a knowledge entry by ID
Get a single knowledge entry by its ID
Get all knowledge entries scoped to a specific project directory
List knowledge entries with optional filters and pagination
List all tags used in stored entries
List all technologies referenced in stored entries
Full-text search across all knowledge entries. Returns results ranked by relevance.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Forces agents to guess what data is returned.
Tool descriptions lack context on WHEN to use each tool. E.g., search_entries vs list_entries distinction unclear. No guidance on dependencies (e.g., 'call list_technologies first to see available filters').
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE). Agents cannot distinguish safe from dangerous operations.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Get summary statistics about stored knowledge entries
Store a new knowledge entry in the brain
Update an existing knowledge entry (partial update — only provided fields are changed)
Parameter constraints missing. 'type' field accepts free-form strings instead of enum (learning|project_context|gotcha). 'limit' lacks min/max bounds (spec says default 10, max 50 but not enforced in schema).
No error handling guidance visible in code. Agents cannot recover from failures (e.g., 'entry not found' should suggest search_entries; 'invalid type' should list valid options).