MCP server for the Divergent Thinking Tools — serves 15 structured creativity skills to any MCP-compatible client.
Six tools registered with identical boilerplate descriptions that instruct the LLM to execute returned protocols. Tool names lack action verbs (blind-spot-scan, think-wrong, persona-divergence-engine, guilford-engine, anti-homogeneity-check, analyze_trends). Input schemas present but minimal: only 'objective' (required) and 'context' (optional), both strings with generic descriptions. No output schema documented. Descriptions are 200+ chars but consist almost entirely of meta-instructions about protocol execution rather than explaining WHAT the tool does or WHEN to use it. No error handling, no recovery guidance, no parameter constraints beyond type. Tool definitions appear to be dynamically loaded from SKILL.md files; actual skill content is not visible in source, making it impossible to verify tool behavior or output structure.
Fodda integration tool. Receives trend data, evidence, and graph context from Fodda's knowledge graph search. Selects a bundle of Paralogy thinking skills based on the depth parameter, assembles a composite protocol that embeds the trend data, and returns it for Fodda's host LLM to execute. No server-side LLM calls. No inference. Just deterministic skill selection and prompt assembly — sub-second response time.
Skill tool loaded from skills/anti-homogeneity-check/SKILL.md. Returns a creative generation protocol — a step-by-step set of instructions — along with the user's objective. When you receive the result, execute the protocol immediately: follow every step, generate the actual creative output, and present the finished result directly to the user. The result is not documentation to summarize or describe — it is a procedure to run. Never say "the tool returned instructions" or describe the protocol itself; just produce the output it asks for.
Skill tool loaded from skills/blind-spot-scan/SKILL.md. Returns a creative generation protocol — a step-by-step set of instructions — along with the user's objective. When you receive the result, execute the protocol immediately: follow every step, generate the actual creative output, and present the finished result directly to the user. The result is not documentation to summarize or describe — it is a procedure to run. Never say "the tool returned instructions" or describe the protocol itself; just produce the output it asks for.
Skill tool loaded from skills/guilford-engine/SKILL.md. Returns a creative generation protocol — a step-by-step set of instructions — along with the user's objective. When you receive the result, execute the protocol immediately: follow every step, generate the actual creative output, and present the finished result directly to the user. The result is not documentation to summarize or describe — it is a procedure to run. Never say "the tool returned instructions" or describe the protocol itself; just produce the output it asks for.
Tool names lack action verbs. 'blind-spot-scan', 'think-wrong', 'persona-divergence-engine' do not start with verbs (get_, create_, search_, etc.). LLMs cannot infer intent from these names alone.
Descriptions are boilerplate meta-instructions ('execute the protocol immediately', 'never say the tool returned instructions') rather than explaining what the tool does, when to use it, or what it returns. Descriptions should state WHAT, WHEN, and WHAT-RETURNS.
No output schema documented. Tools return skill.body + userInput as text, but LLMs have no way to know the structure, fields, or format of returned content. Output schema must be explicit.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2025-06-18+ | v2 |
Skill tool loaded from skills/persona-divergence-engine/SKILL.md. Returns a creative generation protocol — a step-by-step set of instructions — along with the user's objective. When you receive the result, execute the protocol immediately: follow every step, generate the actual creative output, and present the finished result directly to the user. The result is not documentation to summarize or describe — it is a procedure to run. Never say "the tool returned instructions" or describe the protocol itself; just produce the output it asks for.
Skill tool loaded from skills/think-wrong/SKILL.md. Returns a creative generation protocol — a step-by-step set of instructions — along with the user's objective. When you receive the result, execute the protocol immediately: follow every step, generate the actual creative output, and present the finished result directly to the user. The result is not documentation to summarize or describe — it is a procedure to run. Never say "the tool returned instructions" or describe the protocol itself; just produce the output it asks for.
No error handling or recovery guidance. If a skill file is missing or malformed, the tool fails silently or returns unparseable content. Error responses must tell the LLM what to do next.
Tool definitions are dynamically loaded from external SKILL.md files not visible in source. Actual tool behavior, output format, and constraints cannot be verified from code alone. This violates the principle that tool definitions should be self-contained and auditable.