socrata-mcp
An MCP server that gives LLM agents typed, cached access to civic open-data portals via Socrata (SODA 2.1 + Discovery API), enabling search, query, profiling, sampling, and CSV export of datasets.
README
socrata-mcp
An MCP server that gives LLM agents typed, cached access to civic open-data portals. Speaks Socrata (SODA 2.1 + Discovery API) today; the provider layer is a thin interface so CKAN can be added later without touching the tool surface.
Highlights:
- Server-side profiling — null rates, distinct counts, min/max, top values computed via aggregate SoQL; the dataset is never downloaded.
- Hard row caps with honest truncation — every query result carries a
truncatedflag; paging uses a stable:idorder. - Disk cache under
~/.socrata-mcp/cachekeyed by query hash, with short TTLs for metadata and configurable TTLs for query results. - Polite by default — request throttling, retries with backoff that honor
Retry-After, optionalSOCRATA_APP_TOKENsent asX-App-Token. - Loud failures — the portal's actual error message is surfaced to the agent, never swallowed.
- Tableau-ready CSV export — streamed, paged download designed to chain
into vizforge's
csv_to_dashboard.
Install
git clone <this repo> && cd socrata-mcp
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
Register with your MCP client (see .mcp.json.example):
{
"mcpServers": {
"socrata": {
"command": "/absolute/path/to/socrata-mcp/.venv/bin/python",
"args": ["-m", "socrata_mcp"],
"env": { "SOCRATA_APP_TOKEN": "optional-app-token" }
}
}
}
Tools
| Tool | What it does |
|---|---|
search_datasets(query, domain?, category?, limit?, offset?) |
Full-text catalog search via the Socrata Discovery API. |
get_dataset(domain, dataset_id) |
Columns with types, row count, update cadence, license, attribution. |
query(domain, dataset_id, …) |
Structured SoQL (select/where/group/order/limit/offset, within_circle, within_box) or raw soql. Validated, paged, row-capped, truncated flag. |
profile_dataset(domain, dataset_id) |
Per-column null rates, distinct counts, min/max for dates/numbers, top values for categoricals — all portal-side. |
sample(domain, dataset_id, n?) |
First n rows (capped at 100) to see real values. |
export_csv(domain, dataset_id, out_path, …) |
Streamed, paged CSV export of any query. |
Example agent flow:
search_datasets("crime", domain="data.seattle.gov")
get_dataset("data.seattle.gov", "tazs-3rd5")
profile_dataset("data.seattle.gov", "tazs-3rd5")
query("data.seattle.gov", "tazs-3rd5",
where="offense_date > '2026-06-10T00:00:00'",
order="offense_date DESC", limit=100)
export_csv("data.seattle.gov", "tazs-3rd5", "out/spd_30d.csv",
where="offense_date > '2026-06-10T00:00:00'")
# → vizforge: csv_to_dashboard("out/spd_30d.csv", ...)
Configuration
All optional, via environment variables:
| Variable | Default | Meaning |
|---|---|---|
SOCRATA_APP_TOKEN |
unset | Sent as X-App-Token; raises portal rate limits. |
SOCRATA_MCP_CACHE_DIR |
~/.socrata-mcp/cache |
Disk cache root. |
SOCRATA_MCP_METADATA_TTL |
300 |
Seconds to cache dataset metadata. |
SOCRATA_MCP_CATALOG_TTL |
300 |
Seconds to cache catalog searches. |
SOCRATA_MCP_QUERY_TTL |
3600 |
Seconds to cache query/profile results (0 disables). |
SOCRATA_MCP_DEFAULT_LIMIT |
100 |
Rows returned when a query gives no limit. |
SOCRATA_MCP_MAX_ROWS |
5000 |
Hard row cap for inline query results. |
SOCRATA_MCP_MAX_EXPORT_ROWS |
1000000 |
Hard row cap for CSV exports. |
SOCRATA_MCP_PAGE_SIZE |
1000 |
Rows fetched per HTTP request. |
SOCRATA_MCP_THROTTLE_INTERVAL |
0.2 |
Minimum seconds between portal requests. |
SOCRATA_MCP_TIMEOUT |
30 |
Per-request timeout in seconds. |
Cache layout: cache/<kind>/<sha256>.json (kind ∈ catalog, metadata, query,
profile), each file {"cached_at": <epoch>, "data": …}. Deleting the directory
is always safe.
Notes:
- Discovery searches use the US endpoint (
api.us.socrata.com); EU-hosted portals are still directly queryable viaget_dataset/queryon their domain. - Raw
soqlexports run as a single request, so give them an explicitLIMIT(default cap 50,000); structured exports page automatically.
Development
.venv/bin/pytest # unit tests (all HTTP mocked)
.venv/bin/pytest -m network # live smoke tests against data.seattle.gov
Architecture: deterministic core (soql.py, cache.py, http_client.py,
profile.py, export.py) with the MCP layer (socrata_mcp/mcp/) as thin
wrappers over a Provider interface (providers/base.py). To add CKAN,
implement Provider in providers/ckan.py — the tool surface stays unchanged.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.