Serves knowledge to any MCP-capable AI client (Claude Code, Cursor, Windsurf, Autobahn, custom agents). Provides persistent, queryable, evolving knowledge layer for AI-assisted coding with two transport options: stdio (default for editor integrations) and HTTP.
14 tools with mostly complete schemas and descriptions, but significant gaps in parameter documentation and error handling guidance. Tool names follow verb_noun convention well (brain_start_session, brain_create_project, etc.). Descriptions average ~80-120 chars, below the 194-char baseline for A+ tools. Most parameters lack detailed constraints (ranges, formats, enums beyond what's shown). No visible output schema documentation. Error recovery guidance absent. Composition is reasonable, tools are single-purpose, but chaining IDs not consistently returned.
Ask the Oracle (chat interface) a question about the brain's knowledge with citations and reasoning
Create a new project for organizing knowledge and sessions
Find a skill by ID or search for skills by name/content
Get the currently active project for the user
Get the user's peer card (hard-override facts about the user's coding style and preferences)
List all projects accessible to the current user
Log a session event (file created, build attempt, user feedback, etc.)
Output schemas not documented. No visible return type specifications for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Parameter constraints missing. 'limit' in brain_retrieve_knowledge has no min/max bounds. 'reasoningLevel' enum is documented but other enums lack descriptions of what each value means. No regex patterns or length limits specified.
Error handling guidance absent. No tool descriptions explain what to do if a call fails (e.g., 'If knowledge not found, try brain_teach_knowledge'). No recovery paths documented.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | <=2025-11-25 | v2 |
Report the outcome of a coding session and submit learnings for the Knowledge Extraction Agent
Mark a knowledge item as retired/obsolete
Retrieve relevant knowledge items based on semantic similarity and filters
Search sessions by prompt text or event payload using full-text search
Start a new coding session and retrieve relevant knowledge based on the task prompt
Manually teach the brain a new knowledge item (rule, recipe, heuristic, principle)
Get information about the current authenticated user and their capabilities
Chaining IDs not guaranteed. brain_start_session returns sessionId, but unclear if brain_report_session_outcome and brain_log_event receive the same ID format. No documentation of ID field names across tools.
Pagination not visible. brain_list_projects and brain_session_search accept 'limit' but no offset/cursor or total_count documented. Large result sets could exhaust context.