Humanitarian Data Exchange (HDX) MCP — data.humdata.org, the open humanitarian-data platform run by UN OCHA on CKAN. Provides access to humanitarian datasets on internal displacement, refugees, food security, conflict events, disasters, health, and population data.
Four well-named, read-only tools with complete input schemas and solid descriptions. All tools follow verb_noun naming (search_, dataset_, list_). Descriptions are 150 - 250 chars, explaining WHAT the tool does and WHEN to use it. Input schemas include types and required fields. However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract fields. Parameter descriptions lack format constraints (e.g., 'rows: 1 - 1000' is stated in description but not as JSON Schema min/max). No error handling guidance. No tool annotations (readOnlyHint, idempotentHint). Missing pagination metadata (total count, next_cursor). Baseline: 194 chars avg description (these are 150 - 250, good); 100% of A+ tools document output schemas (this server: 0%).
Full dataset record by id or slug (CKAN package_show), including its resources. Each resource has a "download_url"/"url" pointing at the actual data file (usually CSV/XLSX) — that is how you read the underlying rows, since HDX does not expose a keyless row query.
List the locations on HDX (CKAN group_list) — countries and crises, each keyed by a lowercase ISO3 code (e.g. {"name":"syr","display_name":"Syrian Arab Republic"}). Use a code as fq="groups:<name>" in search_datasets to scope results to a country/crisis.
List the publishing organizations on HDX (CKAN organization_list) — UN agencies and NGOs such as OCHA, UNHCR, WFP, IDMC, ACAPS. Use an org "name" as fq="organization:<name>" in search_datasets.
Search the Humanitarian Data Exchange catalogue (CKAN package_search) for datasets on displacement, refugees, food security, conflict, disasters, health, population, etc. Returns matching datasets with titles, descriptions, publishing organization, locations (groups), and resources. Filter by country/crisis with fq="groups:<iso3>" (e.g. "groups:syr") — get codes from list_locations.
Output schemas completely undocumented. LLMs cannot infer what fields to expect from search_datasets, dataset_details, list_locations, or list_organizations. This forces agents to guess field names and risks extraction errors.
Numeric parameter constraints (rows: 1 - 1000, limit: 1 - 1000) stated only in descriptions, not in JSON Schema. LLMs cannot read description text reliably for constraints, use JSON Schema min/max properties.
No tool annotations (readOnlyHint, idempotentHint, destructiveHint). All four tools are read-only and idempotent, annotating them helps agents reason about safety and retry logic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
search_datasets and list_* tools lack pagination metadata in descriptions. No mention of total count, next_cursor, or how to handle large result sets. Agents cannot plan multi-page queries.
No error handling guidance. If a search returns 0 results or a dataset_id is invalid, the tool throws a generic error. Descriptions should guide agents on recovery (e.g., 'If no results, try list_locations to find valid country codes').