DatasetPlaza MCP
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
DatasetPlaza MCP
Read-only dataset routing for agents. It does not fetch data, hold data, or resell access. It answers one question: which published source can answer this, how do I reach it, and is the platform actually entitled to claim that?
The catalogue behind it holds 37,847 dataset cards — China National Bureau of Statistics annual and provincial tables, Hong Kong open data, World Bank, Guizhou portals and more. Each card carries a capability profile, a locator with its indicator list, an evidence provenance, an access level and a machine-readable next step. The agent then connects to the source itself.
Why this exists
Search engines and LLM answers happily return a plausible-looking dataset that cannot actually answer the question. This server is built around the opposite promise: say "no" honestly.
NO_MATCH+shelf_gap— no capability statement on the shelf can take this.ACCESS_RESTRICTED+access_gap— a route exists but requires registration, authorisation or payment; the platform never pays or borrows credentials.OK— only when a candidate is genuinely verified with no named gap.
Evidence provenance is explicit: declared (source-side measured statement),
source_mapped (translated from the publisher's own classification tree),
inferred (guessed from a name by us). Only declared enters verified_axes
and can unlock complete. CONTRACT.md is the full contract, including the
evaluation baseline and the known limits.
Related MCP server: FirstData MCP
What this repository contains
Code and contract only. No catalogue. The index is ~313 MB, which exceeds GitHub's per-file limit and is not something a git history should carry. The data snapshot is distributed separately; see Data.
Install
No third-party dependencies. Python 3.10+.
pip install datasetplaza-mcp
# or, one-shot:
uvx datasetplaza-mcp --data /absolute/path/to/catalogueData
--data expects a directory that holds listings/listings.json, plus a thin
index under listings/ that is derived from it. Put the catalogue there and
build the index once:
python scripts/build_shelf.py --no-rowsThen point the server at that directory root:
datasetplaza-mcp --data /absolute/path/to/catalogueMCP client configuration
{
"mcpServers": {
"datasetplaza": {
"command": "uvx",
"args": ["datasetplaza-mcp", "--data", "/absolute/path/to/catalogue"]
}
}
}Transport is stdio JSON-RPC 2.0. A remote Streamable HTTP gateway
(python -m bdsp.gateway) is available for self-hosting; it is anonymous,
read-only by default, IP rate-limited and binds to 127.0.0.1. Add
--strict-readonly to remove the single write tool from the public surface.
Tools (13)
dp_map, dp_catalog, dp_discover, dp_index, dp_card, dp_compare,
dp_hub, dp_hub_route, dp_pitfalls, dp_report_pitfall, dp_usage,
dp_forms, dp_audit.
Every tool carries annotations.readOnlyHint. Twelve are read-only;
dp_report_pitfall appends a "pitfall" note (it never modifies or deletes) and
is rate-limited to 5 writes per day per caller on the public gateway.
Boundaries
Leads are not proof of existence and contain no values.
An undeclared licence is not permission to redistribute.
Metadata goes stale. Cards reflect what was measured at collection time; verify against the source before relying on an access level or an endpoint.
The catalogue metadata remains subject to each publisher's own terms. The code licence does not cover the data.
Licence
Apache-2.0. The patent grant and the termination clause are the reason this
project is not MIT; see LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Find where real data lives: verified registries, APIs and datasets with recipes to pull them.
Search 77k public-sector open data APIs across 10 countries. Read-only, no API key.
Lets your agent find the right MCP server, tool or service for a task, in one call. Ads labelled.
Machine-readable entity discovery with provenance, trust and verified source evidence.
Related MCP Servers
- FlicenseNot gradedqualityAmaintenanceEnables users and AI agents to turn natural-language data questions into vetted authoritative source recommendations, with step-by-step website navigation, API details, and traceable evidence paths to primary data.-
- FlicenseNot gradedqualityAmaintenanceEnables AI agents and users to turn natural-language questions into authoritative primary data sources, recommending the most relevant government, international organization, or research datasets with authority ratings, access URLs, and API details. It also supplies step-by-step website navigation instructions so users can go from a vague query to the exact official data or document they need.-
- FlicenseNot gradedqualityAmaintenanceEnables AI agents and users to turn natural-language questions into precise paths to authoritative primary data sources, recommending the most relevant government, international organization, and research datasets with authority ratings, API details, and traceable evidence chains. It also provides step-by-step website navigation instructions for locating specific documents and data on complex official portals.-
- FlicenseNot gradedqualityAmaintenanceEnables users and AI agents to turn natural-language data questions into authoritative primary-source recommendations with verified URLs, metadata, APIs, and step-by-step website navigation for retrieving the original data.-