Symbolic derivation and verification toolkit for AI agents via MCP. Provides mathematical computation, formula derivation, and Lean certification capabilities.
SymKit MCP has 7 tools with moderate definition quality. Tool naming follows verb_noun convention (math, session_start, session_show, session_complete, session_verify_session, session_certify, formula_search), which is good. However, descriptions are present but brief (avg ~80 chars), and critical parameter documentation is sparse. The 'math' tool accepts 4 parameters but only 'operation' and 'expression' have clear descriptions; 'session' and 'units' lack detail on when/why to use them. Session tools have naming issues: 'session_verify_session' is redundant (verb_noun_noun). Input schemas are visible but incomplete, most parameters lack type constraints (enums, ranges, patterns). Output schemas are not documented in the source. Error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in metadata.
Search the formula catalog by name, domain, category, or full-text query
Unified Mathematica-style interface for 32 mathematical operations including diff, integrate, solve, dsolve, gradient, laplace, matrix operations, and more
Certify derivation steps using Lean 4 + Mathlib backend when available
Mark the current session as complete
Display the current session state and derivation steps
Initialize a new derivation session with full provenance tracking
Verify the current session and return verification summary
Redundant tool naming: 'session_verify_session' violates single-responsibility principle. Should be 'verify_session' or 'check_session_validity'.
Parameter descriptions incomplete. 'math' tool's 'session' (boolean) and 'units' (object) parameters lack guidance on when/why to use them. 'session_start' goal parameter has no format hints. LLMs cannot infer intent from names alone.
No output schemas documented. Tools return results but callers cannot know what fields to expect. Breaks tool chaining, downstream tools cannot extract required IDs/references.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
No error handling guidance. Tools have risk classifications (READ_ONLY, WRITE) but no recovery hints. If 'math' fails on invalid expression, LLM has no guidance on what to try next.
Tool annotations missing. Risk metadata exists (READ_ONLY, WRITE) but not exposed via MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot infer safety properties.