AI and BI Tool Integration via MCP and APIs
AI and BI Tool Integration via MCP and APIs
This domain covers programmatic access to the semantic layer: REST APIs, embedding, and AI agents that query governed MicroStrategy objects through the Model Context Protocol (MCP).
Core rules
- The REST API authenticates through the login endpoint and returns an authentication token that later calls send in the
X-MSTR-AuthTokenheader, along with the project ID. Endpoints expose reports, dossiers, cubes, datasets, and model objects. - Large result sets are paged with limit/offset and may require asynchronous jobs; plan for token expiry and rate limits.
- Embedding uses the Library or Embedding SDK or the MicroStrategy for Office add-ins; the same object ACLs and security filters apply.
- The MCP server exposes semantic-layer objects (metrics, attributes, reports, dossiers) as tools and resources to LLM clients. It is not a replacement for the semantic layer; it is a governed interface to it.
- Integration governance depends on identity: calls should run as the end user so security filters and object ACLs are honored, and only certified objects should be exposed.
Worked example
An LLM agent connected to MCP reports total revenue for all regions to a regional manager. The MCP connection was configured with a shared service account that has broad privileges, so the query executes under that identity and the manager's row-level filter never applies. The fix is per-user OAuth or token passthrough, plus restricting the MCP tool set to certified metrics. Reasoning: authorization is evaluated for the identity executing the query, not for the person reading the answer.
Traps that cost marks
- Using an administrator or service token for convenience.
- Assuming MCP enforces business rules that live only in a report filter.
- Treating formatted display values as the API numeric values.
- Ignoring pagination and assuming one response contains all rows.
- Mixing API versions or endpoints across releases.