TNL — Typed Natural Language: structured English contracts for coding agents. Provides MCP tools for managing TNL (Typed Natural Language) contract files, including retrieving, tracing, proposing, approving, and verifying changes.
This TNL MCP server provides 6 tools with generally clear naming and reasonable schemas. Strengths: tool names are action-oriented (approve_, get_, retrieve, trace, propose, verify); input schemas are properly typed JSON Schema; tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present for at least one tool (approve_tnl_diff). Weaknesses: parameter descriptions are minimal or missing for most tools (only approve_tnl_diff has a detailed description for diff_id); output schemas are NOT documented, LLMs cannot predict response structure; most tool descriptions lack depth about when to use them or what they return; error handling is basic (generic try-catch with no recovery guidance). The server defines tools via side-effect imports (src/mcp/tools/*.ts) with explicit tool registration in a map, so tool definitions are directly visible, not inferred. Average tool score: 62.
Apply a staged proposal: write each tnl/<id>.tnl, regenerate its sidecar, and remove the staging record.
5 of 6 tools (get_impacted, retrieve, trace, propose, verify) have NO visible input schemas or descriptions. Source files are referenced but content not provided. These tools cannot be reliably used by LLMs without documented parameters and behavior.
No output schemas documented for ANY tool. LLMs cannot predict response structure, anticipate required fields for downstream calls, or plan multi-step sequences. Pattern baseline: 100% of A+ tools have documented return types.
Error handling is generic (catch all errors, return isError=true with message). No recovery guidance. Per pattern baseline, error responses must tell the LLM what to do next (e.g., 'User not found. Try search_users().'). Current handler in src/mcp/tools.ts catches exceptions but provides no contextual recovery steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
Parameter descriptions are minimal or absent. Only approve_tnl_diff documents its parameter (diff_id). The remaining 5 tools have no visible parameter documentation. Pattern baseline: 100% of A+ tools have descriptions for ALL parameters.
Tool descriptions are missing for 5 tools. approve_tnl_diff has a description ('Apply a staged proposal...'), but the others have no visible descriptions in source. Without descriptions, LLMs cannot determine WHEN or WHY to select these tools.