Repository avatar
Monitoring
v0.4.0
active

wikidata-mcp-server

io.github.cyanheads/wikidata-mcp-server

Search and fetch Wikidata entities, execute SPARQL queries, and resolve external identifiers.

Documentation

@cyanheads/wikidata-mcp-server

Search and fetch Wikidata entities, execute SPARQL queries, and resolve external identifiers via MCP. STDIO or Streamable HTTP.

7 Tools • 1 Resource • 1 Skill

Version License Docker MCP SDK npm TypeScript Bun

Install in Claude Desktop Install in Cursor Install in VS Code

Framework

Public Hosted Server: https://wikidata.caseyjhand.com/mcp


Overview

Wikidata's knowledge graph. Search and fetch entities, traverse the graph with SPARQL, and resolve external identifiers like DOI, PMID, ORCID, OpenAlex, and IMDb to their Wikidata items. No API key required.

Tools

ToolDescription
wikidata_search_entitiesSearch for items or properties by text query, returning QIDs/PIDs with labels, descriptions, and match metadata
wikidata_get_entityFetch a full entity by QID or PID with optional field and language filtering
wikidata_get_labelsBatch-resolve up to 500 QIDs or PIDs per call to human-readable labels and descriptions
wikidata_get_statementsFetch property claims for an entity with qualifier detail, per-property paging, a best-rank filter, and QID and PID label resolution
wikidata_get_sitelinksFetch Wikipedia and Wikimedia project article URLs for a Wikidata item
wikidata_sparql_queryExecute a SPARQL SELECT query against the Wikidata Query Service
wikidata_resolve_external_idLook up a Wikidata entity by an external identifier (DOI, PubMed ID, ORCID, OpenAlex ID, etc.)

Resources

ResourceDescription
wikidata://entity/{id}Compact markdown summary of a Wikidata entity — labels, English description, instance-of, Wikipedia link, image, and the number of properties with statements

All resource data is also reachable via tools.

The published wikidata-guide covers entity disambiguation, qualified statements, external identifiers, and SPARQL. It is included in plugin installs and served over MCP as skill://wikidata-guide/SKILL.md.

Capability reference

wikidata_search_entities tool

  • Searches labels, aliases, and descriptions; type="item" for people, places, and works, type="property" for P-IDs
  • Takes a Wikidata language code (en, de, pt-br; case-insensitive); limit defaults to 10 and serves up to 500 per page
  • The endpoint returns no total: page by sending each page's nextOffset as offset (0–9,999) until a page returns no results or no nextOffset; only the first 10,000 results are reachable
  • Each hit includes match metadata saying whether it matched a label or alias

wikidata_get_entity tool

  • Q-IDs (e.g. Q76) fetch items and P-IDs (e.g. P31) fetch properties; fields selects labels, descriptions, aliases, statements, or sitelinks, and narrows the upstream fetch
  • id may be bare, an entity URI (http://www.wikidata.org/entity/Q76), or wd:Q76, as on wikidata_get_labels, wikidata_get_statements, and wikidata_get_sitelinks
  • languages filters multilingual maps to the given language codes
  • An entity too large to inline returns an outline of field categories with byte sizes; its retrieval_notice names a fields set that fits and points oversized statements or sitelinks to the tool that can narrow them

wikidata_get_labels tool

  • Up to 500 distinct IDs per call via the MediaWiki wbgetentities API (batched 50 at a time), with up to 50 language codes per request; ids takes up to 10,000 entries, and distinct IDs past the first 500 come back in deferredIds to send in a follow-up call
  • IDs may be bare (Q76), entity URIs (http://www.wikidata.org/entity/Q76), or wd:Q76, the forms SPARQL results carry
  • Reports a found count and notFound IDs for partial results; malformed entries come back in invalid and fail the call (invalid_ids) only when no valid ID remains
  • Meant to follow a SPARQL query: run it, then humanize the QIDs

wikidata_get_statements tool

  • properties limits the fetch to specific P-IDs (omit for all); item QIDs and quantity units in values, qualifiers, and references are resolved to labels unless resolve_labels=false, and every P-ID the statements name is labelled in propertyLabels
  • rank="best" keeps each property's preferred statements, or its normal ones when none is preferred, and never deprecated ones: the statements SPARQL's truthy wdt: reads (default "all")
  • Each property returns up to limit statements (default 10, at most 500) from offset, in upstream order; propertyTotals counts every statement a property holds, and nextOffset, present while any property holds more, continues every property at once
  • References stay grouped by source; unknown or absent values come back as somevalue / novalue
  • A page of several properties too large to inline returns an outline of P-IDs with byte sizes, largest first (as of 2026-11-15, unfiltered Q30 carries 475 properties across 1,766 statements); re-call with the same limit, offset, and rank plus properties whose sizes sum to 24,000 or less, as the retrieval_notice suggests. A single property is never outlined: it returns its window

wikidata_get_sitelinks tool

  • Maps site codes (e.g. enwiki) to article titles and URLs; sites filters to specific codes
  • wikis_only=true keeps only Wikipedia editions, those whose URL is on a <subdomain>.wikipedia.org host (drops Commons, Wikispecies, Wikidata, Wiktionary, Wikiquote, Wikisource, and every other project)
  • Q-IDs only; P-IDs are not supported

wikidata_sparql_query tool

  • Runs SELECT queries against the Wikidata Query Service (Blazegraph); ASK, CONSTRUCT, DESCRIBE, updates, and plain search text are rejected before the request is sent
  • Standard prefixes (wd:, wdt:, p:, ps:, pq:, wikibase:, bd:) and the wikibase:label SERVICE (when the query uses a ?<var>Label, ?<var>Description, or ?<var>AltLabel variable) are added automatically; language takes one code or a fallback list like fr,en
  • Scholarly articles and their triples (DOI, PMID, authorship) live in a separate graph, reachable only inside SERVICE wdsubgraph:scholarly_articles { … } (wdsubgraph: is predeclared); read an article's label inside that block with BIND(?paperLabel AS ?paperLabel) and a label SERVICE there
  • Results are SPARQL 1.1 JSON bindings, at most 4,000 rows per call; a larger result returns its first 4,000 with truncated, shown, cap, and a notice to narrow the WHERE clause or aggregate, or, for a result not much larger, to page with ORDER BY plus LIMIT/OFFSET (totalCount appears only when the whole result was read). A result body past 8 MiB fails as result_too_large
  • Client timeout is 1–55s (server hard limit 60s); the Query Service allows each client (user agent and IP) 60 seconds of query processing time per 60 seconds and each IP at most 5 parallel queries

wikidata_resolve_external_id tool

  • Resolves DOI (P356), PMID (P698), ORCID (P496), OpenAlex ID (P10283), IMDb ID (P345), and other external IDs to a QID; whitespace and resolver URL prefixes are stripped, and DOIs, PMIDs, ORCIDs, OpenAlex IDs, and IMDb IDs are normalized to their stored form
  • The property must have data type external-id; item-valued or media properties (e.g. P31, P18) are rejected
  • Searches the main graph and the scholarly-article graph (where scholarly articles and their DOIs and PMIDs live) in parallel; a graph that does not answer is named in unavailableGraphs, and the result covers the other
  • Matches through statements of every rank, entities holding the ID at best rank first, then another non-deprecated rank, then deprecated; match.rank reports the statement's rank (deprecated usually means a retired or mistaken ID)
  • Returns match=null when not found (the Query Service lags the live wiki, so null is not proof of absence) and multipleMatches (at most 5) when more than one entity claims the same ID; the label uses the requested language, falling back to mul

wikidata://entity/{id} resource

  • Compact markdown summary — English label and description, the number of label languages with a sample of five, instance-of (P31), enwiki Wikipedia link, image (P18), and the number of properties with statements
  • Returned as text/markdown; for full entity data, use wikidata_get_entity
  • id is a Q-ID or P-ID, the same identifiers the tools take

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

Wikidata-specific:

  • Wikidata REST API v1 for entity and statement fetches — no SPARQL overhead for lookup operations
  • MediaWiki wbgetentities API for efficient batch label resolution
  • Wikidata Query Service (Blazegraph) for SPARQL with auto-injected prefix headers and label SERVICE
  • A count input above its per-call cap is served at the cap with a notice; WIKIDATA_CLAMP_LIMITS=false rejects it instead
  • Configurable User-Agent per Wikimedia policy (WIKIDATA_USER_AGENT)
  • Separate timeout configuration for REST and SPARQL endpoints (WIKIDATA_REST_TIMEOUT_MS, WIKIDATA_SPARQL_TIMEOUT_MS)

Agent-friendly output:

  • All tools echo relevant input parameters in the response for traceability
  • wikidata_get_labels designed to follow SPARQL result sets — run the query, then humanize in one call
  • wikidata_resolve_external_id returns multipleMatches instead of a false null when a Wikidata data-integrity issue puts duplicate claims on one external ID
  • Oversized entities and statement sets return a kind: "outline" payload — field/property sizes plus a retrieval_notice naming the exact re-call — instead of a truncated or oversized response

Getting started

Connect to the hosted instance, no install:

claude mcp add --transport http wikidata-mcp-server https://wikidata.caseyjhand.com/mcp

Or run it locally:

claude mcp add wikidata-mcp-server -- npx -y @cyanheads/wikidata-mcp-server

For other clients, add this to the MCP config file:

{
  "mcpServers": {
    "wikidata-mcp-server": {
      "command": "npx",
      "args": ["-y", "@cyanheads/wikidata-mcp-server"]
    }
  }
}

For Docker or a source build, see Running the server.

Configuration

All configuration is validated at startup via Zod schemas. Key environment variables:

VariableDescriptionDefault
MCP_TRANSPORT_TYPETransport: stdio or httpstdio
MCP_HTTP_PORTHTTP server port3010
MCP_HTTP_ENDPOINT_PATHHTTP endpoint path where the MCP server is mounted/mcp
MCP_PUBLIC_URLPublic origin override for TLS-terminating reverse-proxy deploymentsnone
MCP_AUTH_MODEAuthentication: none, jwt, or oauthnone
MCP_SESSION_MODEHTTP session mode: stateful, stateless, or autostateless
MCP_LOG_LEVELLog level (debug, info, notice, warning, error)info
MCP_GC_PRESSURE_INTERVAL_MSOpt-in forced-GC interval in ms (Bun only)0 (disabled)
LOGS_DIRDirectory for log files (Node.js only)<project-root>/logs
STORAGE_PROVIDER_TYPEStorage backend: in-memory, filesystem, supabase, cloudflare-kv/r2/d1in-memory
WIKIDATA_USER_AGENTUser-Agent string for Wikimedia requests (policy requires a descriptive value)wikidata-mcp-server/0.1 (https://github.com/cyanheads/wikidata-mcp-server)
WIKIDATA_SPARQL_TIMEOUT_MSMax wait for a SPARQL response in ms, retries included; caps every call's timeout (at most 55000)55000
WIKIDATA_REST_TIMEOUT_MSMax wait for REST API responses in ms10000
WIKIDATA_CLAMP_LIMITSServe a count above its per-call cap at the cap, with a notice; false rejects it insteadtrue
OTEL_ENABLEDEnable OpenTelemetryfalse

See .env.example for the full list of optional overrides.

Running the server

Local development

Requires Bun v1.4.4 or higher (or Node.js ≥24.0.0).

git clone https://github.com/cyanheads/wikidata-mcp-server.git
cd wikidata-mcp-server
bun install
cp .env.example .env
  • Build and run the production version:

    # One-time build
    bun run rebuild
    
    # Run the built server
    bun run start:http
    # or
    bun run start:stdio
    
  • Run checks and tests:

    bun run devcheck  # Lints, formats, type-checks, and more
    bun run test      # Runs the test suite
    

Docker

docker run --rm -p 3010:3010 ghcr.io/cyanheads/wikidata-mcp-server:latest

The image defaults to HTTP transport, stateless session mode, and logs to /var/log/wikidata-mcp-server. For stdio client configuration, use "command": "docker" with "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/wikidata-mcp-server:latest"]. Build locally with docker build -t wikidata-mcp-server .; OpenTelemetry peers are installed by default (--build-arg OTEL_ENABLED=false omits them).

Project structure

DirectoryPurpose
src/index.tscreateApp() entry point — registers tools and resources, inits services.
src/mcp-server/toolsTool definitions (*.tool.ts). Seven tools for entity lookup, statements, sitelinks, SPARQL, and external ID resolution.
src/mcp-server/resourcesResource definitions. Entity summary resource.
src/services/wikidataWikidata service layer — REST API client, SPARQL client, statement normalization, types.
src/configServer-specific environment variable parsing and validation with Zod.
tests/Unit and integration tests, mirroring the src/ structure.
skills/wikidata-guide/Published guide for entity resolution, statements, and SPARQL workflows.

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

  • Handlers throw, framework catches — no try/catch in tool logic
  • Use ctx.log for logging, ctx.state for storage
  • Register new tools and resources in the createApp() arrays
  • Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.