Loading ToolBench
Self-hosted MCP search and controlled sharing for agent session history
Brain MCP has well-structured tool definitions with complete schemas and descriptions for all 14 tools. Naming follows verb_noun conventions (search, browse, read_session, prepare_upload, etc.). All parameters include type definitions and descriptions. Tool annotations are properly declared (READ_ONLY, WRITE, DELETE). However, descriptions are often generic and lack actionable context for LLM selection. Parameter descriptions rarely explain constraints, formats, or when to use related tools. Output schemas are not documented, LLMs cannot predict response structure. Error handling guidance is absent. Security-sensitive operations (upload, commit) lack confirmation patterns or dry-run support.
Get information about the current user's access level and permissions
Get aggregated statistics about requests for administrative review (admin only)
List request logs for administrative review (admin only)
Browse agent session history with optional filters for person, repository, source, and date range
Cancel an in-progress upload
Commit a prepared upload to finalize the session history ingestion
Output schemas not documented. LLMs cannot predict response structure, forcing them to guess what fields are returned and how to chain tools. E.g., search() returns hits but structure is undocumented.
Parameter descriptions lack constraint details. E.g., 'limit' parameters state range (1-50) but descriptions do not explain what happens if omitted or why the range exists. Format constraints (ISO timestamp for 'since') are not documented in descriptions.
Destructive/write operations lack confirmation or dry-run patterns. commit_upload and cancel_upload are irreversible but have no confirmation step. Agents may accidentally finalize uploads or cancel in-progress operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
List uploads belonging to the authenticated user
Check which blob hashes are missing from the server
Plan an upload by providing session fingerprints to determine what needs to be uploaded
Prepare an upload of session history with specified scope and archive
Read the full content of a specific agent session by ID or UUID
Search agent session history with optional filters for person, repository, source, date range, and roles
Get statistics about the brain server and session history
Get the status of an in-progress upload
Tool descriptions are generic and do not explain when to use one tool vs. another. E.g., search vs. browse both filter sessions but the distinction is unclear. No guidance on dependencies (e.g., 'call browse first to discover repositories, then search within them').
Error handling guidance absent. No tool descriptions explain what errors are possible, how to recover, or what the LLM should do next. E.g., prepare_upload could fail if archiveSha256 is invalid, but no recovery guidance is provided.