
OpenMetadata MCP Server: Setup for Claude, Cursor and Other Apps
Lets Claude, Cursor and other assistants search your OpenMetadata data catalog, trace lineage and check data quality.
OpenMetadata includes this server, so it runs on your own OpenMetadata deployment at /mcp. Add that address to your assistant and sign in with the same SSO or password you use for OpenMetadata. The assistant then acts with your permissions.
Most tools only read. Four can create or change assets, lineage and tests, and they take effect immediately.
Good for
- Data analysts who need to find the right table and understand it
- Data engineers tracing lineage and root causes of broken data
- Data stewards who tag, describe and govern assets
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Supports hosted MCP servers.
Client does not support this transport.
Client does not support this transport.
Install OpenMetadata MCP
Claude Code
- 1
Copy this prompt.
Add the OpenMetadata MCP server to Claude Code. Run this command: claude mcp add --transport http openmetadata https://metadata.example.com/mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires An OpenMetadata deployment with the MCP application on.
Replace metadata.example.com with your own OpenMetadata host, including the port if it is not standard, for example metadata.example.com:8585. The address comes from the server's listing; the client-specific settings are generated and not documented by OpenMetadata.
Built from the package data we hold. Not yet tested by us.
Tools
Searches data assets by keyword
Keyword search for tables, dashboards, tests and other assets by name, owner, tag, tier, service or column name. It can read many known assets in one call.
Try: “Find tables owned by the finance team tagged PII.”
Read ·
search_metadataFinds data by meaning
Finds data assets that match a plain-language description, even when no keywords match. It needs vector embeddings set up on the server.
Try: “Find tables about customer spending behavior.”
Read ·
semantic_searchOpens one or more assets
Returns full details of an asset or up to 10 at once, including custom properties and a certification with its expiry resolved.
Try: “Show details for the orders table.”
Read ·
get_entity_detailsShows upstream and downstream lineage
Returns a compact lineage graph for an asset, at table level by default, or for one column. Transformation SQL is included only if asked.
Try: “What feeds the revenue dashboard?”
Read ·
get_entity_lineageTraces data quality failures
Looks upstream for failing tests that may explain a problem on an asset, then shows the downstream assets affected.
Try: “Why is the daily_sales table failing its tests?”
Read ·
root_cause_analysisLists data quality test types
Lists the test definitions you can use to create a test case on a table or column.
Try: “What tests can I add to a column?”
Read ·
get_test_definitionsFinds business definitions
Searches glossary terms, metrics and Context Center articles for a business question and returns definitions and candidate assets.
Try: “What does monthly active user mean here?”
Read ·
find_contextResolves a glossary term or metric
Returns a glossary term or metric with its definition, the assets and columns it is bound to, sample values and related metrics. Not in the 2.0.7 release; it is on the main branch.
Try: “Which columns hold the net revenue metric?”
Read ·
get_concept_contextReads Context Center notes
Finds or reads question-and-answer notes extracted from files in the Context Center. Searching needs vector embeddings on the server.
Try: “What does our data retention policy note say?”
Read ·
company_contextShows who you are in OpenMetadata
Returns your identity, teams, roles, domains and the assets you own or follow. It cannot read another user.
Try: “What do I own in OpenMetadata?”
Read ·
get_user_contextReads a persona's AI context
Returns the shared AI context document for a persona, in parts, with the persona admin's guidance if one is set.
Try: “Load the context for the analyst persona.”
Read ·
get_persona_contextShows fields for a new asset
Lists the fields, types and allowed values the create tool accepts for one asset type.
Try: “What do I need to create a glossary term?”
Read ·
describe_entity_typeCreates a new asset
Creates a new entity such as a glossary term. It never changes an existing one and fails if the name is taken.
Try: “Create a glossary term called Churn under Finance.”
Write ·
create_entityEdits an existing asset
Updates an existing asset using a JSON Patch, for example owners, tags or descriptions. It takes effect immediately and can overwrite existing values.
Try: “Set the owner of the orders table to the data team.”
Write ·
patch_entityAdds lineage between two assets
Creates a lineage link between two assets.
Try: “Link raw_orders to the orders table as its source.”
Write ·
create_lineageAdds a data quality test
Creates a test case on a table or column from a test definition. It takes effect immediately.
Try: “Add a not-null test on orders.customer_id.”
Write ·
create_test_caseQueries the knowledge graph
Runs a read-only SPARQL query over the OpenMetadata knowledge graph. Non-admins get SELECT only and need the ExecuteSparqlQuery permission. Not in the 2.0.7 release; it is on the main branch.
Try: “List column lineage paths into the revenue table.”
Read ·
sparql_queryShows what is near an asset
Returns the assets within 1 to 3 hops of an asset in the knowledge graph, such as columns, tags, owners and lineage. Not in the 2.0.7 release; it is on the main branch.
Try: “What is connected to the customers table?”
Read ·
entity_neighborhoodFinds everything with a tag
Finds all assets with a classification tag or glossary term. It needs the ExecuteSparqlQuery permission. Not in the 2.0.7 release; it is on the main branch.
Try: “Show me everything tagged PII.Sensitive.”
Read ·
find_by_tagChecks graph data rules
Validates one asset's graph data, or all of it, against SHACL rules. Admin only. Not in the 2.0.7 release; it is on the main branch.
Try: “Validate the orders table's graph data.”
Read ·
shacl_validateDescribes the data ontology
Returns the OpenMetadata ontology, or the definition of one class or property. Not in the 2.0.7 release; it is on the main branch.
Try: “What classes does the ontology define?”
Read ·
ontology_describe
From the official docs
Permissions
medium risk · 3 things to watch
Medium risk
The assistant can read your catalog and, with the right role, edit metadata, lineage and tests. Sign in with a role that can only read unless you want it to make changes.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It can edit assets and add lineage and tests
Four tools create or change metadata, and their descriptions say writes take effect immediately and can overwrite values. Sign in with a read-only role if you only want search. github.com
It acts with your permissions
Whatever your account can see and change, the assistant can too. Use a limited account for experiments. github.com
Some tools need extra permissions
The graph query tools need the ExecuteSparqlQuery permission, and the validation tool is admin only. github.com
Security advisories
No published security advisories for the MCP server as of Nov 1, 2026.
10 other advisories are about the wider OpenMetadata project, not this MCP server. See them on GitHub
Technical details
| Turning on read-only | No read-only switch is documented. Use an OpenMetadata account whose role only allows reading. |
|---|---|
| Talks to | Your OpenMetadata deployment |
| Sign-in and keys | You sign in through OAuth with PKCE against your OpenMetadata SSO or basic auth. Access tokens last 10 minutes and refresh tokens end 30 days after sign-in, per the README. |
| Scopes |
|
Questions
What is the OpenMetadata MCP server URL?
It is your own deployment's address followed by /mcp, for example https://metadata.example.com/mcp. Include the port if it is not standard, such as :8585.
How do I sign in?
Your assistant opens a browser and uses your OpenMetadata single sign-in, or your username and password if basic auth is on. There is no API key.
Can it change my catalog?
Yes. Four tools create or edit assets, lineage and tests. They run with the permissions of the account you signed in with.
Can I run it read-only?
Not with a switch. Sign in with an account whose role can only read, and the write tools will fail.
Do I need to install anything?
No, if your OpenMetadata deployment has the MCP application on. The server is part of OpenMetadata and nothing runs on your computer.
Why do semantic search and Context Center notes not work?
Both need vector embeddings configured on your OpenMetadata server.
Is it official?
Yes. It is built into OpenMetadata and its listing calls it the official OpenMetadata MCP.
Why do I see fewer than 21 tools?
Your OpenMetadata version is older. The 2.0.7 release has 15 tools. The six graph and glossary tools (get_concept_context, sparql_query, entity_neighborhood, find_by_tag, shacl_validate, ontology_describe) are on the main branch only.
Information
updated Nov 2, 2026
- Publisher
- Official
- Transport
- Streamable HTTP
- Auth
- OAuth
- Version
- 1.1.1
- License
- Apache-2.0
- Last commit
- Oct 31, 2026
- Open issues
- 714
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 3Hide
- github.com · github.com
- github.com · github.com
- github.com · github.com