Loading ToolBench
Federated gateway for UNHCR data MCPs - aggregates responses from domain MCPs (Statistics, RDF, IATI) into a single entry point for all UNHCR open data
Server has 11 tools with complete input schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (country_snapshot, funding_gap_analysis, generate_country_brief). Descriptions are present but often generic, many lack specificity about WHEN to use the tool or what distinguishes it from similar tools. Parameters are typed (string, integer) but lack constraints (enums, ranges, patterns). Output schemas are not documented in the visible code. Error handling is absent, no guidance on retryability, user-fixable errors, or recovery paths. Tools like compare_countries and generate_comparison_brief appear to duplicate functionality without clear differentiation. No tool annotations (readOnlyHint, destructiveHint) despite all being READ_ONLY.
Compare two countries across population, indicators, and funding.
Generate a comprehensive snapshot for a country. Aggregates data from Statistics MCP (population), RDF MCP (indicators), and IATI MCP (funding) into a single, normalized response.
Calculate the funding gap for a country in a specific year. Computes: Total Budget - Total Contributions. Also calculates: Gap per refugee, Gap per capita
Calculate funding metrics per refugee for a country.
Generate a narrative comparison brief for two countries.
Generate a narrative briefing note for a country. Creates an agent-ready narrative with: Situation Overview, Population, Funding Situation, Key Indicators, Major Risks, Key Statistics
Output schemas not documented. Tools return data but LLMs cannot see what fields to expect, forcing them to guess downstream field names and breaking tool chaining.
Parameter constraints missing. 'year' accepts any integer with no range validation (e.g., year 9999 or -1000). 'page_size' in search_datasets has no min/max bounds. Unbounded parameters let LLMs pass absurd values.
No error handling guidance. Tools have no documented error cases, recovery paths, or actionable error messages. If a country is not found, the LLM has no guidance on what to try next.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | <=2025-11-25 | v2 |
Generate a narrative briefing note for an operation.
Get the complete dataset catalog organized by domain.
List all available datasets from all domain MCPs.
Generate a comprehensive snapshot for an operation.
Search datasets across all domain MCPs.
Duplicate/overlapping tools without clear differentiation. 'compare_countries' vs 'generate_comparison_brief' both compare two countries. 'country_snapshot' vs 'generate_country_brief' both summarize a country. LLMs will struggle to pick the right one.
Tool annotations missing. All 11 tools are READ_ONLY but lack readOnlyHint annotation. This prevents LLMs from optimizing caching and retry logic.