MCP reference

Connect AI exposes the Query MCP server. Agents query any connected source with SQL. Developer Edition adds a second server, the Management MCP, which lets agents manage Connect AI itself (list, create, test, and delete connections) from any MCP-aware client.

Query MCP: https://mcp.cloud.cdata.com/mcp
Management MCP: https://mcp.cloud.cdata.com/mcp/mgmt

Authentication

Both servers use HTTP Basic auth. Generate a Personal Access Token from Settings → Access Tokens in cloud.cdata.com, then base64-encode the email:PAT pair:

echo -n "[email protected]:YOUR_PAT" | base64

The resulting string is what goes after Basic in the Authorization header. Every client below uses the same encoded credential.

Client setup

Each MCP client has its own way of accepting a remote server URL with an auth header. The endpoint, auth, and tools are identical. Only the registration step differs.

Claude Code

Use the built-in claude mcp add command. No file editing required.

claude mcp add --transport http connectmcp https://mcp.cloud.cdata.com/mcp \
  --header "Authorization: Basic YOUR_BASE64"

For both servers, register a second entry pointing to /mcp/mgmt with a different name (e.g. connectmcp-mgmt).

Cursor

Open Cursor SettingsTools & MCPNew MCP Server. The mcp.json file opens. Paste:

{
  "mcpServers": {
    "cdata-connect-ai": {
      "url": "https://mcp.cloud.cdata.com/mcp",
      "headers": {
        "Authorization": "Basic YOUR_BASE64"
      }
    },
    "cdata-connect-ai-mgmt": {
      "url": "https://mcp.cloud.cdata.com/mcp/mgmt",
      "headers": {
        "Authorization": "Basic YOUR_BASE64"
      }
    }
  }
}

Claude Desktop

Claude Desktop accepts only stdio-based MCP servers, so the remote HTTP server is bridged via the mcp-remote npm package. Edit claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "cdata-connect-ai": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.cloud.cdata.com/mcp",
        "--header",
        "Authorization: Basic YOUR_BASE64"
      ]
    }
  }
}

Requires Node.js installed. Fully quit and restart Claude Desktop after saving.

Other clients

The n8n, Windsurf, Cline, and Continue setups all use the same Basic-auth header against the same URLs. Only the config UI differs. See the per-client guides in the Connect AI AI Tools docs.

The built-in CData Connect AI connector in claude.ai (the chat interface, not Claude Code) is a separate product path that requires a paid Connect AI tier. Developer Edition users should use Claude Code or one of the other clients above.

Query MCP tools

Query and inspect any connected data source. Available on all editions.

Tool Description Key parameters
getCatalogsList available data source connections
getSchemasGet schemas for a specific catalogcatalog
getTablesGet tables in a schemacatalog, schema
getColumnsGet column metadata for a tablecatalog, schema, table
queryDataExecute SQL queries against connected data sourcesquery
getProceduresList stored procedures for a catalog/schemacatalog, schema
getProcedureParametersGet parameter details for a procedurecatalog, schema, procedure
executeProcedureExecute a stored procedure with parameterscatalog, schema, procedure, parameters

Management MCP tools

Developer edition only

Manage connections programmatically from any MCP client. Use these tools to list, create, test, and delete connections without touching the Connect AI UI.

Tool Description Parameters Returns
list_connections List the data source connections already saved to your account Array of connection objects with id, name, and source type
list_available_sources List every data source you can connect to (Salesforce, Snowflake, Jira, etc.) Array with internal source name, display name, version, and category
get_source_properties Describe the configuration fields a given source requires: credentials, URLs, auth method, and advanced options. Call this before create_connection. source: internal source name from list_available_sources (e.g. "Salesforce") Object with basic and advanced field descriptors, including allowed values and conditional follow-up fields
create_connection Save a new connection. For OAuth sources, the response includes a sign-in URL the user must open to complete setup. name, source, properties (field map from get_source_properties) Created connection object including new id; OAuth sources also return a sign-in URL
test_connection Verify a saved connection still works. Use after create_connection, after an OAuth sign-in, or when a connection has started failing. id: from list_connections Success message, or a specific error the user can act on
delete_connection Permanently delete a saved connection. Irreversible. Confirm with the user before calling. id: from list_connections Success acknowledgement, or an error if the id is not found

Pattern: Self-configuring agent

Developer edition only

With both MCP servers connected, an agent can set up a new data source and immediately start querying it. No UI required. The full flow, driven entirely by natural language:

  1. Agent calls list_available_sources to confirm the source is supported
  2. Agent calls get_source_properties to discover required fields
  3. Agent calls create_connection, for OAuth sources, returns a sign-in URL
  4. User opens the URL and completes OAuth in their browser
  5. Agent calls test_connection to verify the connection is live
  6. Agent calls getCatalogs on the Query MCP and begins querying
Example prompts that drive this pattern in Claude Desktop:
What data sources can I connect to?
Show me the required properties for connecting to Salesforce.
Create a new Salesforce connection named Sales Prod using OAuth.
Test the Sales Prod connection and tell me if anything is failing.
Now query the top 10 open opportunities from Sales Prod.
Full reference

For complete parameter definitions, response schemas, and error codes, see the Connect AI MCP documentation.