Loading ToolBench
An MCP server that generates Mermaid diagrams from code and validates Mermaid syntax
The server provides two tools with complete input schemas and reasonable descriptions. However, several quality issues prevent a higher score: (1) Tool naming is acceptable but not idiomatic, 'generate_diagram' and 'validate_mermaid' start with verbs but lack consistent patterns; (2) Descriptions are present but brief (44-73 chars), below the 50-200 char LLM-optimized range; (3) Parameter descriptions are minimal, many lack actionable guidance on constraints or format; (4) Output schemas are undocumented, no description of what fields are returned, forcing LLMs to reason blindly; (5) Error handling is absent from visible code, no guidance on how to recover from Mermaid CLI failures or invalid input; (6) The tools do not compose, generate_diagram writes to disk but there is no read tool to retrieve the output; (7) One parameter (backgroundColor) accepts free-form strings with only a hint in description ('hex color, named color, or transparent'), inviting invalid values instead of an enum or regex pattern. These gaps are typical of community servers scoring in the C range (60-69).
Generate a Mermaid diagram from code and return it as an SVG image
Validate Mermaid diagram syntax and provide feedback
Output schema not documented. Neither tool describes what it returns. generate_diagram likely returns a file path or SVG data, but this is invisible to the LLM, forcing it to guess downstream behavior.
Parameter descriptions lack actionable constraints. 'backgroundColor' accepts 'hex color, named color, or transparent' but no enum or regex pattern, LLM will hallucinate invalid color names. 'mermaid_code' description is purely descriptive with no guidance on size limits, syntax rules, or error recovery.
No error recovery guidance. generate_diagram will fail silently (or crash) if Mermaid CLI is not installed, if the code is invalid, or if file write fails. No tool description tells the LLM what to do when this happens.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 49 | 2026-07-28+ | v2 |
Tool descriptions are too brief (44-73 chars). LLM-optimized descriptions are 50-200 chars and should explain WHAT, WHEN, and WHAT IT RETURNS. Current descriptions lack context on when to call each tool or what the output enables.
Tools do not compose. generate_diagram writes to disk but there is no read or retrieve tool. An LLM cannot fetch the generated SVG back without additional external tools. Composition is broken.
Insufficient input validation visible in code. The handle_call_tool function is truncated, but no explicit validation of backgroundColor hex codes, theme enum values, or dimension ranges is visible. LLM will pass invalid values expecting graceful rejection.
validate_mermaid returns unclear result. Description says 'provide feedback' but no description of what success looks like (JSON? Plain text? Error codes?) or what fields are in the response.