IMF MCP defines 3 tools with complete input schemas and reasonable descriptions. All tools follow verb_noun naming (get_*, search_*). Descriptions are 150-250 chars, within the 10-1024 baseline. However, output schemas are completely undocumented, LLMs cannot predict what fields to expect from API responses. Parameter descriptions are present but lack constraints (enums, ranges, formats). No error handling guidance. No tool annotations (readOnlyHint, etc.). The server is functional but lacks production-grade polish.
Fetch IMF time-series data for an indicator and country. Uses SDMX CompactData format. Example: get_data({ database_id: "IFS", frequency: "A", indicator: "NGDP_XDC", country: "US", start: "2018", end: "2023" }) returns annual nominal GDP for the US.
List all available IMF databases/datasets. Returns database IDs and names you can use with get_data and search_indicators. Example: call with no arguments to browse available datasets like IFS (International Financial Statistics), BOP (Balance of Payments), etc.
Search or list indicator codes available in an IMF database. Returns the code list (dimensions) for a dataset. Example: search_indicators({ database_id: "IFS", query: "GDP" }) finds GDP-related indicator codes.
Output schemas completely undocumented. LLMs cannot predict response structure (fields, types, nesting). Responses are raw IMF SDMX JSON, verbose, unstructured, and difficult for agents to parse.
Parameter descriptions lack format constraints. 'frequency' accepts 'A', 'Q', 'M' but no enum declared, LLM may hallucinate invalid values like 'annual' or 'W'. 'country' accepts ISO 2-letter codes but no pattern/enum enforces it.
No error handling guidance. If IMF API returns 404 (dataset not found) or 400 (invalid indicator), callTool throws generic 'IMF API error: {status}' with no recovery hint. LLM cannot self-correct or suggest alternatives.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
No tool annotations. All three tools are read-only (safe to retry), but readOnlyHint is missing. Agents cannot distinguish safe tools from destructive ones without explicit hints.
search_indicators returns full DataStructure JSON with query hint, but no pagination or result limit. Large datasets may return thousands of indicators, bloating context and degrading LLM reasoning.