MCP server setup
SelectZero Platform exposes a Model Context Protocol (MCP) server, allowing AI assistants to query the data catalog. The server uses Streamable HTTP transport at /api/v1/mcp with Bearer token authentication and works with any MCP-compatible client.
See Available tools below for the full read-only tool list.
Prerequisites
The MCP server is enabled per instance under Settings → General → Modules → Enable MCP. The toggle requires an application restart to take effect.
A personal access token (PAT) is required for authentication. The MCP tools authenticate as the linked user, so the user must have the View catalog privilege. A service API key is refused with HTTP 403 and a message asking for a PAT. A connection or test suite name the user cannot see is reported as not found, the same as a name that does not exist.
Generating a personal access token
PATs are user-bound tokens; the token authenticates as a specific SelectZero Platform user, and tool access respects that user's role and connection permissions.
The user creating the token needs the Generate PATs privilege (configured per role in Settings → Users & Roles).
As that user, open Settings → Profile → PATs.
Press New token, enter a descriptive name and expiry. Regular users can pick from 1, 7, 14, or 30 days, capped by Settings → General → Modules → Max PAT lifetime (defaults to 30 days). Admins issuing their own PATs can additionally pick 90 days, 1 year, or no expiry.
The token (starts with
dqm_pat_) is shown once on creation - copy it immediately.
Every user with the privilege manages their own tokens on that PATs sub-tab of their profile. Admins can view and delete another user's tokens in the Personal access tokens section of that user's edit dialog under Settings → Users & Roles; only the owner can create or regenerate a token.
Connecting your MCP client
The server uses Streamable HTTP transport at /api/v1/mcp with Bearer token authentication.
Pick the tab that matches your client; Other clients covers anything not listed.
Register the MCP server from the terminal. Run this from the project folder where you use Claude Code, as the server is registered per project by default (see Claude Code MCP documentation):
claude mcp add --transport http selectzero-catalog http://<selectzero-host>/api/v1/mcpOpen the generated config file (location is printed in the output) and add the authorization header:
"selectzero-catalog": { "type": "http", "url": "http://<selectzero-host>/api/v1/mcp", "headers": { "Authorization": "Bearer <your-personal-access-token>" } }Restart Claude Code. The catalog tools become available and Claude can query the data catalog conversationally, scoped to the token-owner's role and allowed connections.
VS Code added MCP support starting with version 1.99. GitHub Copilot Chat in Agent mode can use registered MCP servers.
Create a workspace MCP config file at
.vscode/mcp.jsonin the project folder (or add to your usersettings.jsonunder"mcp"for a global config):{ "servers": { "selectzero-catalog": { "type": "http", "url": "http://<selectzero-host>/api/v1/mcp", "headers": { "Authorization": "Bearer <your-personal-access-token>" } } } }Open the Copilot Chat view, switch to Agent mode, and the SelectZero tools appear in the tool picker.
The token is stored in the config file; treat the file as a secret and exclude it from source control. VS Code also supports
${input:variableName}placeholders that prompt for the token on first use - see VS Code MCP documentation for the input syntax.
Open Cursor Settings -> MCP and press Add new MCP server, or edit
~/.cursor/mcp.jsondirectly:{ "mcpServers": { "selectzero-catalog": { "url": "http://<selectzero-host>/api/v1/mcp", "headers": { "Authorization": "Bearer <your-personal-access-token>" } } } }Toggle the server on in the MCP settings panel. Tools appear in the Composer/Chat tool picker.
Microsoft 365 Copilot reaches the SelectZero MCP server via a Copilot Studio agent. End users chat with the agent inside the regular Microsoft 365 Copilot interface; the agent owner does the one-time setup in Copilot Studio.
Prerequisites:
A Microsoft Copilot Studio license and access to Copilot Studio.
The SelectZero MCP endpoint reachable from Microsoft's cloud (or via your tenant's connectivity solution, e.g. on-premises data gateway / Power Platform virtual network).
A SelectZero PAT for the user identity the agent should run as.
Steps (UI labels current as of Copilot Studio 2026; see the Microsoft Learn guide for the authoritative walkthrough):
In Copilot Studio, open an agent (or create a new one). Make sure generative orchestration is enabled for the agent (required for MCP tools).
Open the Tools page, press Add a tool, then New tool, and select Model Context Protocol. The MCP onboarding wizard appears.
Fill in:
Server name - e.g.
SelectZero catalogServer description - a short description so the orchestrator knows when to invoke it (e.g. "Query the SelectZero data catalog: tables, columns, lineage, glossary, test cases.")
Server URL -
http://<selectzero-host>/api/v1/mcp
Under authentication, select API key:
Type -
HeaderHeader name -
Authorization
Select Create. In the Add tool dialog, choose Create a new connection and provide the API key value as
Bearer <your-personal-access-token>(include the literalBearerprefix). Then Add to agent.Publish the agent. End users select the agent inside Microsoft 365 Copilot chat and the SelectZero catalog tools become available conversationally.
The agent's MCP tool inherits the role and connection access of whichever user's PAT the connection was created with. If you need different users to see different slices of the catalog, configure the authentication differently (each user provides their own PAT via OAuth, or use a service-style PAT scoped narrowly).
SelectZero ships a Claude Desktop extension as a .mcpb package. The install dialog takes a SelectZero URL and a PAT; the extension handles stdio↔HTTP bridging internally. The mcp-remote JSON-config bridge below remains supported for environments that prefer it.
Install the extension
Download
selectzero.mcpb(SHA-256).Double-click the file. Claude Desktop opens an Install SelectZero? dialog showing the extension name, icon, and permissions.
Confirm the install. A small form appears:
SelectZero URL - your instance, e.g.
https://selectzero.acme.corp. Must use HTTPS.Personal Access Token - the PAT from Generating a personal access token. Stored in your OS keychain; not written to disk in plaintext.
Click Install. The catalog tools appear in Claude Desktop's tool picker immediately.
To change the URL or PAT later, open Settings → Extensions → SelectZero, edit the fields, Save, then toggle the extension Off → On so the underlying Node process picks up the new values.
How it works
The extension ships a small Node-based stdio↔HTTP proxy. Claude Desktop spawns it as a child process; the proxy forwards MCP messages between Claude Desktop (stdio) and your SelectZero instance (/api/v1/mcp over HTTPS), attaching Authorization: Bearer <your PAT> to every upstream request. Traffic runs directly between the workstation Claude Desktop is on and your SelectZero instance.
Troubleshooting
The extension writes startup and error messages to Claude Desktop's standard MCP log:
macOS:
~/Library/Logs/Claude/mcp-server-selectzero.logWindows:
C:\Users\<you>\AppData\Roaming\Claude\logs\mcp-server-selectzero.log
Look for lines prefixed with [selectzero]. Common failure shapes and what they mean:
Log line | Likely cause | Fix |
|---|---|---|
| URL entered as | Edit settings, use |
| DNS failure | Verify the hostname; try opening the URL in a browser |
| OS trust store missing the TLS root CA | Install your CA into the OS trust store, then toggle Off → On |
| PAT rejected | Generate a new PAT |
| User lacks View catalog privilege | Add the privilege to the user's role |
Alternative: mcp-remote bridge (fallback)
If your environment cannot install desktop extensions (locked-down corporate device management, kiosk mode, etc.), the community mcp-remote bridge is a JSON-config alternative. Functionally equivalent to the extension; just needs npx on your PATH.
Open the Claude Desktop config file:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
C:\Users\<you>\AppData\Roaming\Claude\claude_desktop_config.json
Add the SelectZero MCP server entry:
{ "mcpServers": { "selectzero-catalog": { "command": "npx", "args": [ "-y", "mcp-remote", "https://<selectzero-host>/api/v1/mcp", "--header", "Authorization:Bearer <your-personal-access-token>" ] } } }Restart Claude Desktop. The catalog tools appear in the conversation tool picker.
Any MCP-compatible client following the MCP specification should work against /api/v1/mcp with a Authorization: Bearer <PAT> header. Refer to the client's documentation for the exact config file location and syntax.
If your client does not support Streamable HTTP natively, the mcp-remote bridge shown in the Claude Desktop tab works generically for any stdio-only MCP client.
Available tools
The MCP server exposes the following tools. Apart from the AI data steward group, which writes suggestions and automations, they only read. The AI client receives a full schema for each tool via the MCP protocol; this list is just an at-a-glance overview. An administrator can switch each tool off under Settings → General → Modules → MCP tools (see Modules); a tool switched off is no longer offered to MCP clients after a restart.
Catalog discovery
searchCatalogObjects - find tables/views by name, description, or filters (connection, schema, dataset, custom fields).
getCatalogColumns - column metadata for one object (data type, glossary terms, custom fields).
getColumnLineage - direct (one-hop) column-level lineage for one object.
getObjectDependencies - multi-hop object-level lineage with hop count and direction.
Glossary & custom fields
searchGlossaryTerms - find glossary terms by name or description; returns linked columns.
discoverCustomFields - list configured custom fields, their descriptions, entity types, and option values.
Tests, executions, audit
searchTestCases - find test cases by connection, target object, test suite, severity, or custom fields.
getExecutionHistory - chronological execution history for a test case, with optional time-range and status filters.
getAuditHistory - version history for an entity (catalog object, term, column, test case).
getAuditSnapshot - JSON snapshot of an entity at a specific audit event.
Documentation
searchDocs - searches the SelectZero product documentation and returns matching sections with their page links, for questions about using SelectZero itself.
Analytical SQL (off by default)
generateAnalyticalSql - turns a business question into a validated
SELECTusing catalog metadata. Sends metadata to the configured AI provider and needs the AI assistant; it reads no data rows.executeAnalyticalSql - runs a single read-only
SELECTagainst a connection and returns the rows. Anything that is not a plainSELECTis refused before it reaches the database, and the user needs the View sample data privilege. No AI provider is involved.
Switch either one on under Settings → General → Modules → MCP tools.
Suggestions (require additional configuration)
suggestProfiling - rule-based test case suggestions for a profiled object.
suggestTestCases - AI-powered test case suggestions (off by default: switch on suggestTestCases under Settings → General → Modules → MCP tools; needs the AI assistant enabled).
AI data steward tool group - submit field suggestions (on a catalog target or on a glossary term) and test case suggestions into the Automation queue, read the decisions people made and why, list, create, update and delete automation rules (always created without a schedule; scheduling is a separate update), size a scope. The group can be disabled under Settings → General → Modules. Accepting and denying suggestions is not available over MCP by design.
Recipes
Multi-step playbooks are exposed as MCP Resources for AI clients that support them:
selectzero://recipes/root-cause-analysis- chain test discovery, execution history, audit history, and lineage to trace a failing check.selectzero://recipes/coverage-gaps- find untagged, undocumented, or untested catalog assets.selectzero://recipes/impact-analysis- find downstream consumers of an object and the quality checks that protect them.