US state & local government contracts + spending, normalized across jurisdictions — keyless
Server defines 7 tools with consistent naming (verb_noun pattern: contracts_by_*, spend_by_*, list_*) and reasonable descriptions (avg ~100 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are completely undocumented, LLMs cannot predict what fields to expect from search results. Parameter descriptions lack format constraints (e.g., no enum for jurisdiction keys despite hardcoded list). Error handling is absent from visible code. No tool annotations (readOnlyHint, idempotentHint). Descriptions are adequate but generic; they don't explain when to use contracts_by_vendor vs contracts_by_keyword, or what the normalized output shape looks like.
Search awarded contracts by agency/department name across all configured jurisdictions
Free-text search for awarded contracts by keyword (title, description, vendor) across all configured jurisdictions
Search awarded contracts by vendor name across all configured jurisdictions
List all available contract jurisdictions with their metadata (platform, level, source URL)
List all available spend/checkbook jurisdictions with their metadata (platform, level, source URL)
Search government spending records by agency/department name across all configured spend jurisdictions
Search government spending records by vendor name across all configured spend jurisdictions
Output schemas completely undocumented. LLMs cannot predict what fields search results contain (vendor, amount, dates, etc.). Forces agents to guess or fail on downstream processing.
jurisdiction parameter accepts free-form strings but only specific keys are valid (tx, wa, king-county-wa, etc.). Should be declared as enum to prevent hallucinated jurisdiction codes.
No error handling guidance. Code does not show what happens when jurisdiction is invalid, API fails, or no results found. LLMs need recovery hints (e.g., 'Try list_jurisdictions() to see valid keys').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
No tool annotations. All tools are read-only but lack readOnlyHint. Agents cannot distinguish safe tools from destructive ones without explicit hints.
Descriptions do not explain the difference between contracts_by_vendor, contracts_by_agency, and contracts_by_keyword. When should an agent choose keyword search over vendor search? Ambiguous tool selection.