ontario-data-mcp 0.2.0

MCP server for searching, downloading, and analyzing datasets from Ontario, Toronto, and Ottawa open data portals

ontario-data-mcp

[!IMPORTANT]
Beta: This project is under active development. The data structure and tool interfaces may change. LLM-generated analysis may contain errors. Always verify critical findings against the returned source data.

This is an MCP server for discovering, downloading, querying, and analyzing datasets from Ontario's Open Data portals. It allows asking questions of the data in English (or Spanish, Chinese, French, etc).

It currently supports the Ontario, Toronto, and Ottawa portals, and utilizes a shared DuckDB cache for fast SQL queries, statistical analysis, and geospatial operations.

Contributing

Contributions welcome! To get started, see Installation below.

Found a bug? Have an idea? Discovered something interesting? Open an issue here: https://github.com/sprine/ontario-data-mcp/issues

Features

• find - search across supported Ontario open data portals
• download - retrieve and cache datasets
• query - run SQL, statistical, and geospatial analysis via DuckDB
• validate — verify that data claims are supported by query results
• A shared DuckDB cache for high-performance analytics

Portal APIs find → Dataset download → DuckDB cache → MCP tools (find, download, query)

Installation

With Claude Code

claude mcp add ontario-data -- uvx ontario-data-mcp

To auto-approve all tool calls (no confirmation prompts), add to your Claude Code settings:

{
  "permissions": {
    "allow": ["mcp:ontario-data:*"]
  }
}

Tools are annotated as read-only or destructive per the MCP spec. Download tools populate the local cache but are read-only (no remote mutations). Destructive tools (cache_manage, refresh_cache) only modify local cached data.

With VS Code

Add to .vscode/mcp.json:

{
  "mcpServers": {
    "ontario-data": {
      "command": "uvx",
      "args": ["ontario-data-mcp"]
    }
  }
}

From Source

git clone https://github.com/sprine/ontario-data-mcp
cd ontario-data-mcp
uv sync
uv run ontario-data-mcp

To connect from source to Claude Code:

Note: MCP subprocesses don't inherit your shell's PATH, so you must use the absolute path to uv (find it with which uv).

claude mcp add ontario-data -- /absolute/path/to/uv run --directory /path/to/ontario-data-mcp ontario-data-mcp

Supported Portals

All searches fan out to every portal by default — no need to select a portal. Dataset and resource IDs are prefixed with their portal (e.g. toronto:abc123).

Portal	Platform	Datasets
ontario	CKAN	~5,700
toronto	CKAN	~533
ottawa	ArcGIS Hub	~665

List of tools available to the AI agent

Discovery (5 tools)

Tool	Description
search_datasets	Search for datasets across all portals (or narrow with portal=)
list_portals	List all available portals with platform type
list_organizations	List government ministries with dataset counts
list_topics	List all tags/topics in the catalogue
find_related_datasets	Find datasets related by tags and organization

Metadata (4 tools)

Tool	Description
get_dataset_info	Get full metadata for a dataset (use prefixed ID like toronto:abc123)
list_resources	List all files in a dataset with formats and sizes
get_resource_schema	Get column schema and sample values for a datastore resource
compare_datasets	Compare metadata side-by-side for multiple datasets (cross-portal)

Retrieval & Caching (4 tools)

Tool	Description
download_resource	Download a resource and cache it in DuckDB (use prefixed ID like toronto:abc123)
cache_info	Cache statistics + list all cached datasets with staleness
cache_manage	Remove a single cached resource or clear the entire cache
refresh_cache	Re-download cached resources with latest data

Querying (4 tools)

Tool	Description
query_resource	Query a resource via CKAN Datastore API (remote)
sql_query	Run SQL against the CKAN Datastore (remote)
query_cached	Run SQL against locally cached data in DuckDB
preview_data	Quick preview of first N rows of a resource

Data Quality (3 tools)

Tool	Description
check_freshness	Check if a dataset is current vs. its update schedule
profile_data	Statistical profile using DuckDB SUMMARIZE
validate_result	Validate that a data claim is supported by query results

Geospatial (3 tools)

Tool	Description
load_geodata	Cache a geospatial resource (SHP, KML, GeoJSON) into DuckDB
spatial_query	Run spatial queries against cached geospatial data
list_geo_datasets	Find datasets containing geospatial resources

MCP Resources

Resources the agent can read for context without calling a tool:

URI	Description
ontario://cache/index	List of all locally cached datasets with freshness info
ontario://dataset/{dataset_id}	Full metadata for a specific dataset (supports prefixed IDs)
ontario://portal/stats	Overview statistics across all data portals
ontario://schema/{table_name}	Column schema, types, sample values, and type warnings for a cached table
ontario://guides/duckdb-sql	DuckDB SQL reference for Ontario open data analysis

Prompts

Context-aware guided workflow prompts:

• explore_topic — Guided exploration of a topic (fetches live catalogue context)
• data_investigation — Deep dive into a specific dataset: schema, quality, statistics
• compare_data — Side-by-side analysis of multiple datasets

Environment Variables

Variable	Default	Purpose
ONTARIO_DATA_CACHE_DIR	~/.cache/ontario-data	DuckDB storage + log file location
ONTARIO_DATA_TIMEOUT	30	HTTP timeout in seconds
ONTARIO_DATA_RATE_LIMIT	10	Max CKAN requests per second

Development

uv sync
uv run python -m pytest tests/ -v

License

MIT — see LICENSE for the software.

Data accessed through this tool is provided under the following open government licences:

• Contains information licensed under the Open Government Licence – Ontario.
• Contains information licensed under the Open Government Licence – Toronto.
• Contains information licensed under the Open Government Licence – City of Ottawa.