meta-data-mcp 2.1.1

MCP server exposing 60+ open data providers to LLM clients via intelligent metadata routing.

meta-data-mcp

A single MCP server that transparently routes user requests to 66 open-data sources.

meta-data-mcp is one MCP server — not many. Under the hood it bundles 66 plugins, each wrapping a different open-data API. The plugins are an implementation detail; from your LLM's perspective there is one server, one set of tools, and one place to ask "where can I find data about X?"

You install one server. You get all the data, discoverable through built-in routing tools.

Why "meta"?

Finding open data isn't the hard part — there's an absurd amount of it available. The hard part is finding the right dataset when you need it. meta-data-mcp makes that automatic:

• The LLM calls opendata-find-providers ("FX rates", "court rulings", "earthquakes near Lisbon") and the server routes the query against an internal registry of every bundled plugin.
• The LLM then calls the matching tool directly. No setup step in between, no separate servers, no per-provider install rituals.

This project was forked from opendata-mcp and reshaped around the single-server idea once the catalogue passed a few dozen plugins.

Installation

You'll need uv (a Python package manager).

# macOS — install uv via Homebrew so MCP clients can find it
brew install uv

# Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Then register the server with every MCP client installed on your machine:

uv run meta-data-mcp setup

The command auto-detects which MCP clients you have installed and adds one meta-data-mcp entry under mcpServers in each. Supported clients:

Client	Config file
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json (macOS) / %APPDATA%/Claude/claude_desktop_config.json (Windows)
Claude Code	~/.claude.json
Cursor	~/.cursor/mcp.json
Windsurf	~/.codeium/windsurf/mcp_config.json
Gemini CLI	~/.gemini/settings.json
LM Studio	~/.cache/lm-studio/mcp.json

Each existing config is backed up to <file>.bak before writing. Restart the affected client(s) and you'll see one new server with discovery tools + every plugin's tools available under it.

Inspect what's detected / configured on your machine:

uv run meta-data-mcp clients

Target a single client (or write to every supported client regardless of detection):

uv run meta-data-mcp setup --client claude-code
uv run meta-data-mcp setup --client all

If you want to see the JSON snippet without touching any config file (e.g. to paste into a client we don't support yet):

uv run meta-data-mcp setup --print-json

When META_DATA_MCP_AUTH_TOKEN is set, --print-json also surfaces the SSE-client snippet (with the real token) to stderr so you can wire a remote client.

Hosting meta-data-mcp as a remote SSE server

For deploying behind your own domain with bearer-token authentication, see docs/hosting.md. It covers systemd, Caddy/nginx TLS termination, token rotation, and the threat model.

CLI

There is one server, so the CLI takes no "provider" argument. Every command operates on the one meta-data-mcp server.

Command	What it does
uv run meta-data-mcp run	Run the server (default SSE; pass --transport stdio for Claude Desktop).
uv run meta-data-mcp setup	Register the server in Claude Desktop's config.
uv run meta-data-mcp remove	Unregister the server from Claude Desktop.
uv run meta-data-mcp cleanup	Detect and remove legacy multi-server entries (--apply to commit).
uv run meta-data-mcp inspect	Launch mcp-inspector against the server.
uv run meta-data-mcp list	Informational: list the internal plugins bundled in this server.
uv run meta-data-mcp info	Informational: show server overview. Pass --plugin <name> for plugin-level details.
uv run meta-data-mcp version	Print the package version.

The list command exists for transparency about what's bundled — plugins are not separately installable, runnable, or addressable. They are loaded automatically when the server starts.

Server tools (what the LLM calls)

Once meta-data-mcp is running, the LLM has access to two layers of tools — and you don't need to mention either to the user:

1. Meta tools — the eight server-level tools below. They make routing transparent: the LLM uses them to find (or create) the right plugin without you telling it which tool to call.
2. Plugin tools — ~330 tools coming from the 66 bundled plugins. They are merged into the same namespace at startup. The LLM picks one after consulting the meta tools.

Meta tools

Tool	Purpose
opendata-find-providers	Free-text search over the plugin registry. Returns ranked matches. When nothing matches the response carries a no_match: true flag and a next_step hint pointing at opendata-draft-spec + opendata-create-plugin.
opendata-explain-choice	Show the scoring breakdown for a search (useful for debugging routing decisions).
opendata-list-domains	Enumerate the controlled domain vocabulary (health, legal, finance, earth-science, …).
opendata-list-regions	Enumerate the controlled region vocabulary (us, eu, uk, global, …).
opendata-describe-provider	Full metadata for one plugin by id — title, description, domains, regions, keywords, homepage, required env vars.
opendata-list-providers	Paginated dump of the whole registry.
opendata-draft-spec	Build a validated plugin YAML spec from structured inputs. Takes id, base_url, tool definitions (name, endpoint, params), and registry metadata. Validates id/tool-name casing, path-placeholder/param consistency, and parameter types, then emits a YAML string ready to feed into opendata-create-plugin. Use this so the LLM never has to hand-author YAML.
opendata-create-plugin	Autonomously create a new plugin. Takes a YAML spec (typically produced by opendata-draft-spec), runs the generator, imports the new module, registers it in the live registry, and hot-loads its tools onto the running server. Use this when opendata-find-providers returns no match.

The autonomous discovery flow

The reason this server is called "meta" is that it routes data requests on the user's behalf — including by creating the route when one doesn't exist yet. The full flow:

1. User asks for data, e.g. "show me the most recent published CVEs."
2. LLM calls opendata-find-providers with the query (cve, vulnerability, …).
3. If the registry has a match: the LLM picks the matching plugin's tool and answers — done.
4. If the registry has no match: the response includes no_match: true and a next_step field that explains the autonomous creation path. The LLM:
   1. Tells the user it's about to add coverage for this data source.
   2. Web-searches for an open API that exposes the requested data (e.g. the NVD or CIRCL CVE API).
   3. Calls opendata-draft-spec with the API's id, base URL, and structured tool definitions. The server validates the inputs (id casing, path-placeholder consistency, parameter types) and returns a YAML string.
   4. Passes that YAML to opendata-create-plugin. The server materializes the plugin module + tests, imports the module, registers a ProviderEntry in the in-memory dynamic registry, and merges the new tools into the running server's tool list.
   5. Calls the newly-available tool to answer the user's original question.
5. User gets their answer — and the plugin remains available for the rest of the session.

The materialized plugin lives on disk (meta_data_mcp/providers/{id}.py + tests/providers/test_{id}.py); contributors can clean it up, add it to meta_data_mcp/registry.py as a static entry, and open a PR so it becomes part of every shipped install.

Plugin tools

Every bundled plugin contributes its own tools under the one server. Their names are unique kebab-case identifiers, often using a provider-specific prefix (e.g. usgs-eq-feed-significant-week, frankfurter-latest, wikipedia-fetch-summary). The LLM discovers them through opendata-find-providers and opendata-describe-provider; you don't need to memorize them.

Presentation layer (MCP Apps)

v2.0 adds a visual layer on top of every tool result. Hosts that support the MCP Apps extension (Claude Desktop, MCP Inspector, others) render bound tool results inline as interactive panels in a sandboxed iframe instead of as JSON text. Hosts that don't speak MCP Apps fall back to the same JSON they always got — the binding is purely additive.

Each MCP-Apps-aware tool declares its panel via _meta.ui.resourceUri on the tool description. The host fetches the ui:// resource (HTML + bundled JS, single payload, no external requests besides explicitly-whitelisted CDNs) and dispatches bidirectional postMessage events between the iframe and itself.

Shape primitives — ui://meta-data-mcp/shape/<name>/v1

Three reusable bundles cover the common payload contracts. Any tool whose response matches one of these shapes binds to the corresponding primitive automatically and gets a rich renderer for free.

Shape	Renders	Payload contract
timeseries/v1	Line chart + auto-computed profile (min/max/mean/stddev/gap-count) via Plotly.	{points: [{date, value, series?}], axes: {x, y}, annotations?}
geofeatures/v1	Leaflet map + marker cluster (with density layer for high-cardinality outputs).	`{features: GeoJSON
records/v1	Faceted, sortable, paginated HTML table + per-column auto-profile (type inference, top-k, null rate, range).	{rows: [...], schema?, default_facets?}

Custom apps — ui://meta-data-mcp/app/<name>/v1

Some data shapes don't fit a generic primitive. v2.0 ships dedicated apps for them:

App	Drives	Visualization
discovery/v1	opendata-find-providers, opendata-list-domains, opendata-list-regions, opendata-activate-provider, etc.	Faceted plugin browser with live health badges.
vulnerability/v1	nvd-*, osv-*, epss-*, cisa-kev.	CVSS radar + severity heatmap + exploitation-probability gauge.
entity-graph/v1	crossref-works-by-author, openalex-search-works, wikidata-search-entities, opensanctions-search.	Force-directed graph (D3) with co-authorship overlay.
trade-flows/v1	comtrade-trade-data.	Reporter → commodity → partner Sankey + commodity treemap.
news-tone/v1	gdelt-article-search, gdelt-volume-timeline.	Volume + tone timeline with country-pair chord diagram.
network-topology/v1	ripestat-asn-neighbours and friends.	Force-directed ASN peering/upstream/downstream graph.
molecular/v1	pubchem-compound, pdb-entry.	WebGL 3D structure viewer (3Dmol.js, cartoon for proteins, stick+sphere for ligands).
museum/v1	met-search, met-search-by-artist, met-get-object.	Lazy-loaded CSS-grid image gallery + provenance detail panel.

Building new apps

Adding a UI binding to a generated provider is now a one-line spec change:

tools:
  - name: my-tool
    description: ...
    endpoint: /foo
    response_shape: records   # ← binds to the shape primitive

See tools/specs/README.md for the full reference. Bundle-size budgets are enforced in CI (warn ≥ 100 KB, error ≥ 1 MB); the v2.0 bundles range from 14 KB (timeseries primitive) to 34 KB (vulnerability app), all comfortably inside the budget.

Bundled plugins (66)

This is what's inside the one server. You don't install these individually — they all come along.

Government / Civic

Plugin	Source	Description
au_data_gov	Australian Government Open Data	CKAN catalog at data.gov.au
ca_open_gov	Canada Open Data	CKAN catalog at open.canada.ca
fr_data_gouv	data.gouv.fr	French government open data platform
nl_tweedekamer	Tweede Kamer	Dutch Parliament open data
sg_data_gov	Singapore Open Data	data.gov.sg datasets and collections
uk_gov	data.gov.uk	UK government CKAN catalog
us_cary	Town of Cary Open Data	Town of Cary, NC open data via Socrata — public safety, transportation, utilities, parks
us_data_gov	Data.gov	US federal government open datasets
us_fayetteville	City of Fayetteville Open Data	City of Fayetteville, NC open data via Socrata — public safety, infrastructure, community services
us_raleigh	City of Raleigh Open Data	City of Raleigh open data via Socrata — public safety, infrastructure, parks, planning

Statistics / Economics

Plugin	Source	Description
eu_eurostat	Eurostat	European Union statistics
global_imf	International Monetary Fund	IMF SDMX 2.1 statistical data
global_dbnomics	DBnomics	Global economic data aggregator (IMF, World Bank, etc.)
global_oecd	OECD	OECD economic & social statistics (SDMX)
global_world_bank	World Bank	Development indicators by country
nl_cbs	Statistics Netherlands (CBS)	Dutch statistical datasets (OData v2/v3)
uk_ons	UK ONS	UK Office for National Statistics

Finance / Markets

Plugin	Source	Description
eu_ecb	European Central Bank	ECB data portal (SDMX) — FX, monetary, banking
global_coingecko	CoinGecko	Cryptocurrency market data
global_frankfurter	Frankfurter	ECB reference FX rates (key-less)
us_sec_edgar	SEC EDGAR	Public company filings, XBRL financials
us_treasury_fiscal	US Treasury Fiscal Data	Federal debt, daily Treasury statement, FX rates

Health & Life Sciences

Plugin	Source	Description
global_disease_sh	disease.sh	COVID-19, influenza, vaccine aggregator
global_pubchem	NCBI PubChem	Chemical compounds and substances
global_rcsb_pdb	RCSB PDB	3D protein and macromolecular structures
global_who_gho	WHO GHO	WHO Global Health Observatory (OData)
us_cdc_socrata	US CDC	CDC open data via Socrata
us_clinicaltrials	ClinicalTrials.gov	NIH/NLM clinical trials registry v2
us_fda_openfda	openFDA	FDA adverse events, recalls, labels
us_healthdata_gov	HealthData.gov	HHS open health data via Socrata — outcomes, insurance, demographics, public health

Earth Science / Weather / Environment

Plugin	Source	Description
eu_copernicus	Copernicus (EU)	European Earth observation and climate datasets
global_open_meteo	Open-Meteo	Weather forecast + historical + air quality
us_ncdeq_gis	NC DEQ Environmental GIS	NC Dept. of Environmental Quality ArcGIS Hub — permits, air/water quality, hazardous waste
us_noaa_ncei	NOAA NCEI	Climate data access services (key-less)
us_noaa_tides	NOAA Tides & Currents	Water levels, tides, currents
us_usgs_earthquake	USGS Earthquakes	Real-time and historical seismic events

Biodiversity / Space / Physics

Plugin	Source	Description
cern_opendata	CERN Open Data	Particle physics datasets and software
global_gbif	GBIF	Global biodiversity occurrence records
global_inaturalist	iNaturalist	Citizen-science species observations
global_opensky	OpenSky Network	Live ADS-B flight tracking
us_nasa	NASA	APOD, Near Earth Objects, Mars rover photos

Geo / Mapping / Knowledge

Plugin	Source	Description
global_osm_nominatim	OSM Nominatim	Geocoding / reverse-geocoding (1 req/sec)
global_overpass	OSM Overpass	Query OpenStreetMap with Overpass QL
global_wikidata	Wikidata	Structured knowledge graph + SPARQL
global_wikipedia	Wikipedia	Article summaries, related, page views
us_arcgis_item	ArcGIS REST API	Fetch public ArcGIS item metadata by ID — layers, maps, services, files
us_census_geocoder	US Census Geocoder	Address ⇄ coordinates ⇄ geographies
us_nc_onemap	NC OneMap	NC's authoritative GIS clearinghouse via ArcGIS REST — statewide geographic layers

Transit / Aviation

Plugin	Source	Description
ch_sbb	Swiss Federal Railways	Swiss train disruptions and service data
de_db	Deutsche Bahn	German railway open data
nl_ndov	NDOV Loket	Dutch public transport data
us_faa_nasstatus	FAA NAS Status	US airspace status, delays, ground stops (XML)
us_noaa_awc	NOAA Aviation Weather	METAR, TAF, and station weather data

Scholarly Literature

Plugin	Source	Description
global_arxiv	arXiv	Preprint metadata (Atom XML)
global_crossref	Crossref	DOI metadata, citations, journals
global_europepmc	Europe PMC	Biomedical literature + fulltext XML
global_openalex	OpenAlex	Open scholarly metadata

Culture / Books

Plugin	Source	Description
global_met_museum	Met Museum	Met Museum Open Access (CC0)
global_open_library	Open Library	Books, authors, works (Internet Archive)
global_unesco_heritage	UNESCO World Heritage Sites	Natural, cultural & mixed World Heritage Sites

Networking / Internet

Plugin	Source	Description
global_bgpview	BGPView	BGP routing data — ASN info, prefixes, peers (key-less)
global_ripe_stat	RIPE NCC RIPEstat	Production-grade BGP data (key-less)

Legal

Plugin	Source	Description
nl_rechtspraak	Dutch Rechtspraak	Dutch court rulings and case law (ECLI)
uk_legislation	UK legislation.gov.uk	UK Acts, statutory instruments (XML/Atom)
us_courtlistener	CourtListener	US court opinions, dockets, judges (Free Law Project)
us_federal_register	US Federal Register	Daily rules, notices, executive orders

Optional environment variables

A few bundled plugins accept optional API keys for higher rate limits. Set these in your shell or in the Claude Desktop server config's env block:

Variable	Plugin	Purpose
COURTLISTENER_API_TOKEN	us_courtlistener	Anonymous access works at low volumes
META_DATA_MCP_CONTACT	all	Your email, used in User-Agent for polite-pool APIs (Crossref, OpenAlex, OSM, SEC EDGAR). Defaults to [e-mail].

Server runtime flags

Variable	Purpose
META_DATA_MCP_PRELOAD	Comma-separated plugin ids to activate at startup, or * for all. Default unset = discovery-only (~13 meta tools).
META_DATA_MCP_AUTH_TOKEN	When set on the SSE transport, requires Authorization: Bearer <token> on /sse and /messages.
META_DATA_MCP_PROVENANCE	Truthy (1, true, yes, on) enables a meta-data-mcp/provenance entry on every tool-call result's first content block's _meta, carrying sha256 and timestamp (ISO 8601 UTC, ms precision). The digest covers the canonical (tool, arguments, content) envelope — content blocks dumped via model_dump(mode="json", by_alias=True, exclude_none=True) with _meta stripped, JSON-serialized with sort_keys=True, separators=(",",":"), ensure_ascii=True. Binding tool name + arguments into the hash means audit logs can distinguish "tool A returned X" from "tool B returned X". Default off — opt in when you need tamper-evidence. See meta_data_mcp/provenance.py module docstring for the verbatim receiver recipe.

Transports

run defaults to SSE (HTTP, port 8000) so you can connect from the MCP Inspector or remote clients. For Claude Desktop (which the setup command targets), the spawned process uses stdio:

uv run meta-data-mcp run                                  # SSE on 127.0.0.1:8000
uv run meta-data-mcp run --transport stdio                # stdio
uv run meta-data-mcp run --host 0.0.0.0 --port 3001       # SSE bound to all interfaces

Roadmap

• v1.2 — Hierarchical discovery: opendata-list-subcategories and opendata-browse-providers tools so users can browse domain → subcategory → provider when they don't know what they need.
• v1.3 — Agent-driven generation: Hook the routing engine's no-match path into an agent that finds an open API, generates a provider module, and registers it automatically — closing coverage gaps without user intervention.
• Public hosted deployment with SSE so non-Claude clients can use the server remotely.
• Multi-language SDK clients for the discovery tools so non-MCP integrations get the same routing benefits.

Credits

• Originally conceived by grll as opendata-mcp.
• Forked and reshaped around the single-server "meta-mcp" model.
• Built on Anthropic's open-source MCP spec.

License

MIT — see LICENSE.