Conduit is a federation gateway, not a primary MCP server. The three demo tools (add, multiply, reset) are minimal stubs in docs/demo/downstreams.py, not production tools. Tool definitions lack rigor: 'multiply' has an empty schema object (no properties, no required fields); 'add' and 'reset' have minimal descriptions (under 20 chars); no parameter descriptions; no output schemas documented; no error handling guidance. The 'reset' tool is marked DESTRUCTIVE but has no confirmation pattern or permission gate. These are demo/test fixtures, not real tools, and score accordingly.
Add two integers
Multiply two integers
A deliberately denied demo tool
multiply tool has empty input schema (type:object with no properties or required fields). Cannot validate or guide LLM on expected parameters.
All three tools have descriptions under 20 characters ('Add two integers', 'Multiply two integers', 'A deliberately denied demo tool'). Descriptions must be 10 - 1024 chars and explain WHAT, WHEN, and any prerequisites for LLM selection.
No parameter descriptions. 'a' and 'b' in add() lack context; LLM cannot infer whether they are counts, IDs, or measurements. Every parameter needs a description.
reset tool is marked DESTRUCTIVE but has no confirmation pattern, dry-run option, or permission gate. Destructive operations must support confirmation or require explicit authorization.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract return values without knowing the response structure.