Loading ToolBench
Exposes RecCli's temporal memory engine to MCP-compatible agents (Claude Code, Cursor, Windsurf, etc.) as callable tools. Handles LLM communication, .devsession management, and cross-session context retrieval.
RecCli MCP server has 11 well-named tools with clear verb-noun patterns (list_, get_, search_, rebuild_, validate_). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present with typed parameters and descriptions. However, output schemas are not documented in the source code, critical for LLM planning. Error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Parameter descriptions are adequate but lack format constraints (e.g., date_range format, filter object structure). The server is functional but lacks production-grade polish around output documentation and error recovery.
Retrieve code changes recorded in sessions with file paths, diffs, and rationale.
Load full context around a specific message: the message itself, surrounding conversation, linked summary items, and related decisions/problems.
Aggregate summary across all sessions in a project: timeline, key decisions, architectural patterns, recurring problems, and overall progress.
Load a specific .devsession file and return its full conversation, metadata, and summary.
List all recorded decisions across sessions with their rationale, alternatives considered, and implementation status.
List all recorded problems and their solutions across sessions.
List all recorded .devsession files in the project, with metadata and summaries. Excludes live in-progress snapshots.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what fields does get_session return? Does search_sessions return session_id for chaining?). Critical for tool composition.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications in metadata (READ_ONLY, WRITE, REVERSIBLE). Agents cannot distinguish safe vs. dangerous operations.
Error handling lacks recovery guidance. No documented error responses, retryability classification, or actionable error messages. E.g., what happens if session_id is invalid? What should the LLM do next?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Rebuild the unified vector index from all .devsession files. Useful after adding new sessions or if index becomes corrupted.
Search across all .devsession files using vector embeddings and hybrid retrieval (dense + sparse BM25). Returns ranked results with similarity scores.
Generate or retrieve a summary of a .devsession: overview, key decisions, problems solved, code changes, and next steps.
Validate integrity of .devsession files: checksums, schema compliance, embedding consistency.
Parameter constraints underspecified. 'filters' object in search_sessions lacks schema definition. 'date_range' format not documented. 'file_pattern' glob syntax not explained. LLMs will guess at valid formats.
Pagination parameters present (limit, offset) but no total_count or next_cursor documented in output. LLMs cannot determine if more results exist or how to fetch them efficiently.