Last updated: July 12, 2026

MCP Server

Brickcost exposes a read-only MCP surface for agents that need readiness evidence, Genie risk questions, metric reconciliation drafts, semantic-layer candidates, data-product contracts, and operating graph context.

MCP is the transport, not the product moat. The differentiator is the customer-validated evidence and reconciliation backlog exposed through the tools. Use Databricks managed MCP where native Genie, SQL, Vector Search, or Unity Catalog functions are enough; use the Brickcost MCP surface when agents need the diagnostic verdict, benchmark questions, owner-review semantic drafts, and grounding readiness.

Endpoints

• POST /api/mcp for JSON-RPC MCP calls.
• GET /api/mcp/manifest for deployment metadata and allowed tools.
• GET /api/mcp/tools and POST /api/mcp/tools/{tool} for REST-compatible testing.

Controls

The current MCP runtime accepts a configured bearer token. It does not implement Databricks App browser-session or Databricks OAuth authentication. A static contract check does not validate a deployed endpoint; live authentication and JSON-RPC testing are required before an MCP listing is published.

• BRICKCOST_MCP_BEARER_TOKEN enables bearer-token auth for external agents.
• BRICKCOST_MCP_TOOL_ALLOWLIST limits which read-only tools are exposed.
• BRICKCOST_MCP_RATE_LIMIT_PER_MINUTE enables per-actor request limiting.
• Successful tool calls are written to the product activity log.

Safety

MCP tools are read-only: they return readiness evidence and generated artifacts and do not execute changes in your workspace. Generated SQL remains an owner-review draft that your team applies; Brickcost does not apply it.

BI Readiness Tools

The MCP surface includes tools for customer-backed BI evidence, including find_risky_genie_questions, find_dashboard_consolidation_candidates, and explain_customer_bi_readiness. These tools use the same imported dashboard metadata and operating graph shown in the app.

Connector-aware agents can call get_connector_evidence to retrieve dbt, workflow, ticket, orchestration, and Git evidence already folded into the operating graph. Agents can call generate_optimization_experiment_plan to retrieve a control/treatment canary plan for a materialized view or data-product recommendation. Agents can call generate_data_product_contract to retrieve an owner-review contract draft with freshness SLA, quality checks, access policy, AI-grounding status, and migration plan for a recommended governed data product.