On this page

MCP

Kind: Mcp
Configuration object: mcp

Use MCP to execute one explicitly selected tool. WebIQ uses the application's REST search adapter. Other providers use streamable HTTP MCP, discovering the tool and validating its schema before execution.

Inputs

argumentsJson is JSON-object text whose string values can contain placeholders such as {{input}} or {{question}}. A whole named placeholder preserves its JSON type. Exact {{input}} remains a string.

Standalone steps require an explicit tool name. To let a model choose among discovered tools, configure MCP entries inside Multi-turn.

Configuration

Property	Default	Usage
`provider`	`WebIQ`	`WebIQ`, `XMcp`, `FoundryIQ`, `WorkIQ`, or `FabricIQ`.
`enabled`	`true`	Enables the integration. A disabled standalone step fails if invoked; disabled Multi-turn entries are omitted.
`endpoint`	`""`	Published public HTTPS endpoint. Provider defaults are applied by configuration helpers; invocation requires a resolved endpoint.
`endpointName`	`""`	Optional Vault or host endpoint reference, resolved before execution.
`allowEndpointSecrets`	`false`	Explicit consent for a credential-bearing connection URL.
`credentialName`	`"WEBIQ_API_KEY"`	Runtime API-key or bearer-token reference appropriate for the provider.
`tokenScope`	Unset	Required for IQ providers. Select the audience and permission supported by that endpoint.
`toolName`	`"webiq_search"`	Explicit discovered tool name for a standalone step.
`argumentsJson`	WebIQ query/count/locale template	Object text, at most 256 KiB, matching the selected tool's schema.

Provider setup

Provider	Connection and authorization
`WebIQ`	Default endpoint `https://api.microsoft.ai/v3/search/web`, tool `webiq_search`, and `x-apikey` authentication. Optional token-scope authentication uses a bearer token.
`XMcp`	Default endpoint `https://api.x.com/mcp` and bearer authentication. Only read-only allowlisted tools are accepted, with discovered safety annotations checked.
`FoundryIQ`	Your knowledge-base MCP endpoint and configured Entra identity. The helper suggests `knowledge_base_retrieve` with `https://search.azure.com/.default`; verify the published endpoint and schema.
`WorkIQ`	Administrator-approved remote MCP endpoint. Native Work IQ requires delegated authorization and preconsented permissions; the helper suggests `ask` and `api://workiq.svc.cloud.microsoft/.default`. Graph-backed resource access uses `fetch`, `get_schema`, and `search_paths`.
`FabricIQ`	Native endpoint `https://fabriciq.svc.cloud.microsoft/v1/mcp/fabriciq` with delegated Power BI authorization and `https://analysis.windows.net/powerbi/api/.default`. Requires `Item.Read.All`, `Item.Execute.All`, and `Dataset.Read.All`. Service-principal and application-only authorization are unsupported.

Foundry IQ and Work IQ require resource-specific or administrator-approved endpoints. Native Fabric IQ has a published endpoint and read-only tools for Power BI reports and semantic models. It discovers artifacts, reads metadata and schemas, searches stored values, and executes DAX with RLS and OLS preserved. Ontologies and data agents use separate item-specific integrations with their own endpoints, tools, and scopes. See the Fabric IQ MCP service reference.

The dedicated Work IQ and Fabric IQ toolbox entries create provider-specific MCP configurations. A standalone Fabric IQ step still needs an explicit discovered tool name and matching arguments. The Work IQ and Fabric IQ sample categories configure full research workflows, data-access consent, and human evidence review.

A generic bearer connection cannot supply additional per-user authorization required by some protected knowledge sources. Explicit non-X tool selection can permit data-modifying operations, so review the selected tool and server.

WebIQ arguments are limited to query, count (integer 1-50), language, and region. Query must be nonblank. Locale values contain 2-16 letters or hyphens. Defaults are count 5, language en, and region US.

The X allowlist is get_news, get_posts_by_id, get_posts_by_ids, get_posts_counts_recent, get_posts_liking_users, get_posts_quoted_posts, get_posts_reposted_by, get_trends_by_woeid, get_users_by_id, get_users_by_username, get_users_by_usernames, get_users_posts, get_users_timeline, search_news, search_posts_all, and search_users.

Outputs

WebIQ returns formatted search text containing titles, URLs, and passages, or a no-results message.

For MCP, supported text, resource text, and resource links are combined. A structured-only result returns structuredContent as JSON. When both structured content and supported text exist, the complete tool-result object is returned, preserving both. A declared output schema is validated. Unsupported or empty results and tool errors fail a standalone step.

Example

{
  "id": "search",
  "name": "Search public sources",
  "kind": "Mcp",
  "outputName": "research",
  "mcp": {
    "provider": "WebIQ",
    "enabled": true,
    "endpoint": "https://api.microsoft.ai/v3/search/web",
    "credentialName": "WEBIQ_API_KEY",
    "toolName": "webiq_search",
    "argumentsJson": "{\"query\":\"{{input}}\",\"count\":5,\"language\":\"en\",\"region\":\"US\"}"
  }
}

Supply the WebIQ credential, then connect the output to a Model step using the research binding. MCP discovery and execution occur during a run, so applying settings alone does not contact a tool server.