ca.artsdata由 artsdata.ca 發布官方發布者v1.0.0更新於 Oct 24, 2026
Search Artsdata for authoritative data on Canadian events, artists, organizations, and places.
面向 Artsdata 知識圖譜的遠端 MCP 伺服器，讓助理檢索加拿大活動、藝術家、機構與地點，並執行唯讀 SPARQL 查詢。
Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/ca.artsdata%2Fmcp/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"artsdata": {
"type": "http",
"url": "https://mcp.artsdata.ca/mcp"
}
}
}4.0.68.1POST /mcp: MCP Streamable HTTP endpoint (tools and resources are documented in
docs/mcp-server.md)GET /up: health checkGET /llms.txt: machine-readable guide for agentsCalls to /mcp are reported to Google Analytics 4.
artsdata://dumps/core-minus-provenance/latest (artsdata_dump): manifest for the
latest Artsdata core-minus-provenance dump — the core graph with provenance and
RDF-star annotations removed, published monthly as a gzipped Turtle file.
resources/read asks the Artsdata Databus for the newest version of the artifact
(GET /databus/artifact/latest?artifact=<ARTIFACT_URI>) and returns a small JSON
document with the download URL, version, format and compression — never the dump
contents. Going through the Databus means MCP clients need to know nothing about
where the file is actually stored. Clients download the file from downloadUrl and
compare version across reads to detect when a new dump lands.
ARTSDATA_RECONCILIATION_ENDPOINT: reconciliation service URL used by the
search_entities, get_entity and search_events tools.ARTSDATA_SCHEMA_URL (optional): Turtle file compiled by the get_schema tool
(default https://docs.artsdata.ca/artsdata-schema.ttl).ARTSDATA_SCHEMA_CACHE_TTL_SECONDS (optional): how long the compiled schema is cached
in process (default 86400).ARTSDATA_CORS_ORIGINS (optional): comma-separated CORS origins (default *).ARTSDATA_API_ENDPOINT (optional): base URL of the Artsdata API, used to query the
Databus for the latest data dump (default https://api.artsdata.ca).ARTSDATA_SPARQL_ENDPOINT (optional): SPARQL endpoint queried by the sparql_query
tool (default https://query.artsdata.ca/query).ARTSDATA_SPARQL_TIMEOUT_SECONDS (optional): read timeout for a sparql_query call
(default 25).ARTSDATA_SPARQL_MAX_ROWS (optional): maximum rows sparql_query returns; extra rows
are cut and the result is flagged truncated (default 1000).env/production.envenv/staging.envARTSDATA_INSTANCE_TYPE (optional): PRODUCTION (default) or STAGING.
The Docker entrypoint loads the matching file from env/.Calls to /mcp are reported to Google Analytics 4, including the tool used,
how long the call took, and the caller's User-Agent. Reporting stays off until
both variables below are set, so development and CI need no configuration.
GA4_MEASUREMENT_ID: the GA4 property's G-XXXXXXXXXX id. Not a secret;
defaulted for production in config/environments/production.rb.GA4_API_SECRET: Measurement Protocol secret. Secret — set it as a Heroku.
config var, never in env/*.env, which is committed.GA4_DEBUG (optional): tag events so they appear in GA4's DebugView within
seconds, and log whether GA4 considers each payload valid. Events are still
recorded. Useful because the normal endpoint answers 204 whether or not the
payload is valid.Reporting is on when both credentials are present and off otherwise.
Separately, every MCP call is written to stdout as one JSON line, including the
complete arguments the agent sent — the full SPARQL query, search terms, entity
URIs. On Heroku these lines
go to heroku logs and to any drain you attach.
MCP_REQUEST_LOG (optional): set to false to silence the request log. It
runs independently of Google Analytics and works with no GA4 credentials set.For sparql_query calls, two extra facts are reported: sparql_form
(SELECT, ASK or UNKNOWN) and sparql_where, the
WHERE clause as the agent wrote it with the PREFIX block removed and whitespace
collapsed. Both go to Google Analytics and to the request log; GA4 cuts the
WHERE clause at its 100-character limit, the log keeps it whole.
The API will be available at http://localhost:3000.
Build the image:
Run the server:
Run with staging profile:
Run tests in Docker:
Run a single test file:
Run all tests:
The public/llms.txt file is generated from MCP server metadata (tools, resources, descriptions, and schemas).
CI behavior:
Developer workflow:
This project uses GitHub Actions for continuous integration and automated deployment to Heroku.
Every push to the main branch triggers a workflow that builds and deploys the application to Heroku.
The workflow is defined in .github/workflows/deploy-to-heroku.yml.
Every pull request triggers a workflow that unit tests. The workflow is defined in .github/workflows/run-unit-tests.yml.
The Artsdata MCP Server is actively deployed and hosted in a live production environment.
Live Production Endpoints
來源：README.md，提交 2ed7f52
Tool to search for entities in the Artsdata knowledge graph by name, with optional filters on entity type and on the controlled vocabulary (skos:ConceptScheme) a concept belongs to. Each result has a markdown-link `label` like '[Name (K12-345)](https://kg.artsdata.ca/resource/K12-345)': use it verbatim when mentioning the entity.
limitquery必填typeslanguagein_schemeTool to get entities details from Artsdata Knowledge Graph by entity URI. Entity URIs end in the Artsdata ID (K\d+-\d+, e.g. K12-345); cite it with the entity name. The result's `sources` lists the datasets the entity was built from: name them when you answer about the entity, linking `url` when present, else by `name`.
uri必填Tool to search for events in the Artsdata knowledge graph by place, artist, organization, has_event_type_concept and language. Places, artists and organizations are Artsdata entity URIs; the has_event_type_concept filter takes concept URIs from the Artsdata Event Types vocabulary (adr:ArtsdataEventTypes), not schema.org classes. Resolve any name or label to its URI with search_entities first. Places must be venues; for events in a city or region, use sparql_query (see the places parameter). At least one filter is required. Entity URIs end in the Artsdata ID (K\d+-\d+, e.g. K12-345); cite it with the entity name.
limitplacesartistslanguagestartDateToorganizationsstartDateFromhas_event_type_conceptReturns the data model of the Artsdata Knowledge Graph: its entity classes (Event, Place, Organization, Person, LivePerformanceWork...), the properties of each with required/optional, cardinality, value types, linked classes and allowed values, the controlled vocabularies, the SPARQL prefixes, and conventions for querying. Derived from the Artsdata Ontology and the Artsdata CORE graph SHACL shapes. Call this before writing a SPARQL query to run with sparql_query against https://query.artsdata.ca/query, since Artsdata's model is not general knowledge. Takes no input.
無需輸入參數。
Runs a read-only SPARQL 1.1 SELECT or ASK query against the Artsdata Knowledge Graph (https://query.artsdata.ca/query) and returns the result as SPARQL 1.1 Query Results JSON, exactly as the endpoint returns it: head.vars and results.bindings for SELECT, boolean for ASK. Call get_schema first and build the query from the classes, properties, prefixes and conventions it returns, since Artsdata's model is not general knowledge. Declare every PREFIX you use and always add a LIMIT: at most 1000 rows are returned (the rest is cut and truncated is true; page with OFFSET), and a query gets 25 seconds. If the endpoint rejects the query, its error message is returned so you can fix the query and retry. For the details of an entity URI in the results use get_entity; to find an entity's URI from its name use search_entities. Always SELECT the entity URI next to its label so its Artsdata ID can be cited. Entity URIs end in the Artsdata ID (K\d+-\d+, e.g. K12-345); cite it with the entity name. To find who claims a property value, see the sources guidance in the get_schema conventions.
query必填已滿足適用檢查項的 90%
該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。
請在儲存庫或註冊中心條目中宣告授權條款。
bundle install.github/ workflows/ deploy-to-heroku.yml # Workflow deploying to Heroku run-unit-tests.yml # Reusable workflow running Minitest unit-test-on-pull-request.yml # Workflow running unit tests on pull requestsapp/ resource/ # MCP resources schema/ # JSON schemas of the tools' input and output services/ # Clients for the reconciliation service, SPARQL endpoint and Databus tool/ # MCP toolsconfig/ initializers/mcp_server.rb # MCP server: registered tools, resources and transport initializers/mcp_analytics.rb # Google Analytics 4 reporting configuration routes.rblib/ mcp_analytics/ # Rack middleware reporting MCP usage to Google Analytics 4test/bundle exec rails serverdocker build -t artsdata-mcp-server .docker run --rm -p 3000:3000 artsdata-mcp-serverdocker run --rm -p 3000:3000 -e ARTSDATA_INSTANCE_TYPE=STAGING artsdata-mcp-serverdocker run --rm artsdata-mcp-server bundle exec rails testbundle exec rails test test/tool/search_events_test.rbbundle exec rails testbundle exec rake docs:generate_llmsbundle exec rake docs:check_llms