Ckan MCP Server

What ckan-mcp-server does

ckan-mcp-server gives MCP-compatible clients access to CKAN catalogs and their public APIs. It can search datasets with Solr syntax, rank results using weighted metadata fields, retrieve complete package metadata, and summarize the resources attached to a dataset. Dataset searches can use Boolean expressions, wildcards, fuzzy matching, proximity searches, range filters, facets, sorting, pagination, and date expressions.

The server also covers catalog structure and data access. Agents can list or search organizations, groups, and tags; inspect individual organizations or groups; obtain portal status and catalog statistics; and query CKAN DataStore resources with filters, selected fields, sorting, distinct values, or SQL. A separate SPARQL tool handles read-only SELECT queries against public HTTPS endpoints such as data.europa.eu, DBpedia, and Wikidata. Portal discovery uses the live datashades.info registry, while MQA tools provide quality scores and failing metrics for datasets on dati.gov.it.

How it works

Each CKAN operation accepts a server_url, allowing one configured server to work with different public portals. A typical workflow verifies the portal with ckan_status_show, checks its locale, searches in the portal's language, inspects a selected package, and then queries a resource whose datastore_active value is true. ckan_analyze_datasets can combine discovery with DataStore schema and record-count inspection before a row query.

For recent or historical searches, distinguish content dates from CKAN record timestamps. issued and modified represent publisher content dates when available; metadata_created and metadata_modified describe CKAN record activity and may reflect harvesting on aggregators. The content_recent option applies an issued-date search with a metadata-created fallback.

Setup and configuration

Install the local server with npm and expose it through an MCP client's stdio configuration:

{
  "mcpServers": {
    "ckan": {
      "command": "npx",
      "args": ["@aborruso/ckan-mcp-server@latest"]
    }
  }
}

A hosted streamable-HTTP endpoint is available at https://ckan-mcp-server.andy-pr.workers.dev/mcp, so clients that support remote MCP connections can use it without a local installation. Claude Desktop can configure either transport; Claude Code supports both the hosted HTTP command and the local npx command.

Tools and capabilities

• Search and relevance-rank CKAN datasets, with facets, filters, pagination, and weighted field scoring.
• Retrieve package metadata, resource details, download URLs, tracking data, and DataStore availability.
• Query DataStore tables directly or submit read-only SQL to portals that expose SQL access.
• Browse organizations, groups, tags, portal status, catalog statistics, and registered CKAN portals.
• Inspect DataStore schemas and record counts before selecting fields for a query.
• Run read-only SPARQL SELECT queries over HTTPS with automatic limits when a query has none.

Limitations and notes

Not every CKAN resource has DataStore enabled, and SQL access depends on the target portal. SPARQL accepts only HTTPS endpoints and SELECT queries. Local and hosted use share a 50,000-character tool-output cap. Dataset search returns up to 1,000 results per page, DataStore row queries allow up to 32,000 rows, and automatic SPARQL limits range up to 1,000 rows when the query does not supply its own LIMIT. Oversized responses are truncated; JSON responses include truncation indicators.

The hosted endpoint has a shared quota of 100,000 requests per day. Local Node.js deployments can configure response-body limits with CKAN_MAX_RESPONSE_BYTES and CKAN_MAX_DECOMPRESSED_BYTES. Source-portal probing for unavailable DataStore resources is disabled by default and can issue additional requests when enabled.