Loading ToolBench
BlueKit MCP Server for managing kits, blueprints, walkthroughs, agents, collections, diagrams, and clones
BlueKit MCP server demonstrates moderate tool definition quality with consistent naming and reasonable descriptions, but suffers from significant gaps in output schema documentation, error handling guidance, and parameter validation. Of 14 tools analyzed, 11 have clear action-verb naming patterns (bluekit_agent_*, bluekit_blueprint_*, bluekit_clone_*, bluekit_diagram_*), but output schemas are entirely undocumented, critical for LLM reasoning about downstream chaining. Descriptions range from adequate (100-250 chars for most tools) to problematic (bluekit_ping at 40 chars, well below the 10-1024 char rubric baseline of 194 chars avg). Parameter schemas are well-typed in most cases, but lack actionable constraints (no enums for constrained fields, minimal validation guidance). Error handling is absent from all tool definitions, no recovery guidance, retryability classification, or actionable error messages documented. The bluekit_batchExecute tool shows composition concerns: it attempts to orchestrate multiple operations but lacks documented transactional semantics or partial-failure handling.
Initialize a BlueKit project by linking it to the BlueKit store. Creates ~/.bluekit directory and adds project to SQLite database.
Start the process of creating a new agent. This tool provides instructions for generating an agent with all required metadata (tags, description, capabilities, etc.).
Generate an agent file in the .bluekit directory of the specified project path with the generated content. This should be called AFTER creating the agent content with proper YAML front matter including tags (1-3 descriptive tags), description (clear, concise sentence), and capabilities (exactly 3 bullet points).
Execute multiple tool operations in sequence (single approval)
Generate a blueprint folder in .bluekit/blueprints/ containing blueprint.json and all task files. Tasks are blueprint-specific instructions (not reusable kits). Optionally save to global registry at ~/.bluekit/blueprintRegistry.json. IMPORTANT: Use bluekit_blueprint_planBlueprint first to validate structure. Read the blueprint definition from MCP resources (bluekit://prompts/get-blueprint-definition.md) for context.
No output schemas documented for any tool. LLMs cannot reason about return types, field names, or structure needed for downstream chaining. Baseline shows 100% of A+ tools have documented return types.
No error handling documentation. Tools do not specify retryability, failure modes, or recovery guidance. Pattern requires: error category (retryable/user-fixable/fatal), actionable error messages with violated constraints, and next-step guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 47 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |
Get a specific blueprint by ID including blueprint.json and all task files. Can retrieve from project folder or global registry.
List blueprints from project's .bluekit/blueprints/ directory and/or global registry at ~/.bluekit/blueprintRegistry.json
Plan a blueprint structure by analyzing requirements and ensuring proper layer parallelization. This tool validates layer dependencies and provides warnings/suggestions before generation. Use this BEFORE bluekit_blueprint_generateBlueprint to ensure correct structure.
Validate an application configuration against a blueprint's config schema. Returns validation errors if the config doesn't match the schema.
Create a new project from a registered clone. Clones the git repository at the exact commit, copies files to target location. Use when user says "create project from clone" or "use the bluekit clone".
Register a git repository snapshot as a clone. Automatically detects git URL, current commit, branch, and tags. Saves to project's .bluekit/clones.json file. Use when user says "clone this project" or "save this as a clone".
Start the process of creating a new mermaid diagram. This tool provides instructions for generating a diagram with all required metadata (alias, description, tags).
Generate a mermaid diagram file in the .bluekit/diagrams directory of the specified project path. This should be called AFTER creating the diagram content with proper YAML front matter including alias, description, and tags.
Health check for BlueKit MCP server
bluekit_ping description is only 40 characters ('Health check for BlueKit MCP server'). Rubric hard-rule: descriptions under 20 chars score 0-20; descriptions this short leave LLMs unable to determine utility. Expand to explain when to call it and what it validates.
Constrained parameters lack enum enforcement. Examples: bluekit_blueprint_* tools accept 'layers' objects with no type/structure definition. bluekit_clone_register accepts 'tags' array with no validation. Rubric requires enums for known sets of values to prevent hallucinated inputs.
bluekit_batchExecute orchestrates multiple tool operations but lacks documented transactional semantics, partial failure handling, and rollback guidance. Composition pattern requires clear semantics: are failures atomic, per-task, or cascading? What happens if task N fails after tasks 1-N-1 succeed?
Tool naming mismatch: 'bluekit.init_project' uses dot notation inconsistently with other tools using underscores (bluekit_agent_*, bluekit_blueprint_*, etc.). Inconsistent naming increases LLM confusion when tool density is high.
No confirmation/dry-run pattern for destructive operations. bluekit_agent_generateAgent, bluekit_blueprint_generateBlueprint, bluekit_clone_createProject, and bluekit_diagram_generateDiagram perform file writes without confirmation step. Pattern requires confirm-before-execute or dry-run for irreversible operations.
Parameter descriptions lack actionable constraints. Examples: 'projectPath' repeated across 8+ tools with generic description 'path to the project directory', no guidance on absolute vs relative, validation of existence, or handling if missing. Rubric requires format, range, and allowed-value guidance.
Tool dependency documentation missing. bluekit_agent_generateAgent's description says 'call AFTER creating agent with bluekit_agent_createAgent' but this dependency is not formally stated in schema or parameter descriptions, forcing LLMs to parse text for intent.