Personal knowledge base MCP server and REST API that allows Claude on claude.ai to query a semantic search database containing emails, messages, notes, documents, conversations, calendar events, financial data, and manually added entries.
Second Brain MCP has 8 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (query_, get_, add_, rate_). Descriptions are substantive (100-250 chars typical), exceeding the 10-char minimum. However, several tools lack parameter descriptions for all fields, and output schemas are not formally documented. Error handling is minimal, tools return success/failure booleans without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Composition is sound: tools are single-purpose and chainable (e.g., query_second_brain returns results that feed into add_entry). Security: no credentials exposed as parameters. Overall: solid foundation, but gaps in parameter documentation and error messaging prevent a higher score.
Add a new entry to the brain directly from the conversation.
Get self-improvement analytics and recommendations for the brain. Returns source quality rankings, top query topics, stale data stats, and actionable recommendations for improving brain quality.
Get a full health and stats overview of the Second Brain. Called when the user asks "how's my brain doing?" or similar. Returns collection stats, health trends, source status, active alerts, and usage insights.
Find all entries mentioning a specific person.
Get all entries related to a project in chronological order.
Get entries added to the brain in the last N hours.
Output schemas not formally documented. Tools return dicts with 'results', 'entries', 'success' fields, but LLMs cannot see the structure of nested objects (e.g., what fields are in each entry?). This forces LLMs to guess at downstream field names and risks extraction errors.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). add_entry and rate_response are clearly write operations; query/get tools are read-only. Annotations would help LLMs reason about side effects and retry safety.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Search your Second Brain — his personal knowledge base containing emails, messages, notes, documents, conversations, calendar events, financial data, and manually added entries. Use this to find information Wes has encountered, decisions he's made, conversations he's had, or anything from his personal and professional life.
Rate the quality of a previous query response.
Error handling lacks recovery guidance. add_entry returns {success: false, message: 'Failed to add entry'} with no hint about why it failed or what to retry. rate_response similarly returns bare success/failure. Per pattern:recovery-guide, errors should guide the LLM's next step.
get_brain_status and brain_insights have complex fallback logic (try/except blocks that silently skip missing modules). If a module fails to load, the response omits that section without warning. LLMs cannot distinguish between 'no alerts' and 'alerts module failed'.
query_second_brain accepts source_filter and tag_filter as arrays but does not document valid enum values. LLMs must guess which source types are valid (conversations, emails, messages, documents, notes, decisions, people, projects, ideas, bookmarks, transactions, balances, holdings, calendar, manual are listed in description but not as schema enums).