MCP server exposing economic time series data from FRED (Federal Reserve Economic Data) and news headlines from NewsAPI.org, plus multi-agent orchestration capabilities
Five tools with complete JSON schemas, clear descriptions (100-200 chars), and proper parameter typing. All tools are read-only with appropriate risk classification. Naming follows verb_noun convention (search_*, get_*, compare_*). Descriptions explain WHAT, WHEN, and output shape. However, parameter descriptions lack detail on constraints (e.g., date format validation, frequency enum rationale). Error handling is mentioned in descriptions but not formalized in output schema. No tool annotations (readOnlyHint, idempotentHint) present. Output schemas are documented in descriptions but not in formal JSON Schema format visible in code.
Fetch and align 2–4 FRED series over the same date range for comparison. Capped at 4 series to keep the response bounded.
Get units, frequency, last-updated date, and source notes for a series. Read-only, small response. The 'notes' field is external text and is returned wrapped as untrusted data.
Fetch observations for one known FRED series ID over a required date range. Refuses unbounded ranges. Long ranges are thinned to stay within the session token budget.
Search recent news headlines relevant to a topic. Returns up to 10 headlines (title, source, published date, short snippet) — never full article text, and never a summary or opinion of what they say. Every title and snippet comes back wrapped as untrusted data, the same pattern as a FRED series' notes field.
Search for a FRED series ID from a plain-language description. Returns candidate series (id, title, units, frequency) only — never observations. Use this first when you have a concept ('unemployment rate') but not an ID.
Parameter descriptions lack constraint details. 'start_date' and 'end_date' state format (YYYY-MM-DD) but do not explain validation rules (e.g., 'start_date must be before end_date', 'range cannot exceed 10 years'). LLMs cannot infer these constraints from names alone.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. All five tools are read-only and idempotent, but this is not formally declared in the tool definition. Agents cannot infer safety properties without explicit hints.
Error handling is described in tool descriptions ('Refuses unbounded ranges', 'Capped at 4 series') but not formalized in output schema. No structured error response format visible (e.g., {"error": "code", "message": "...", "recovery": "..."}). LLMs cannot reliably parse recovery guidance.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Output schemas are documented in descriptions but not in formal JSON Schema format. Tools return 'candidate series (id, title, units, frequency)' and 'up to 10 headlines' but no explicit response schema is visible in code. Agents must infer structure from prose.
Pagination not explicitly supported. 'search_news' returns 'up to 10 headlines' but no limit/offset or next_cursor parameters visible. Large result sets could exhaust context window without pagination mechanism.