MCP server for connecting AI assistants to Tacit building digital twin platform
Four tools with clear verb-noun naming (tacit_list_*, tacit_graphql, tacit_timeseries). Descriptions are well-structured (150 - 250 chars), explaining WHAT, WHEN, and dependencies. All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the source code, LLMs cannot see what fields to expect from responses. Parameter descriptions are good but lack explicit format constraints (e.g., ISO 8601 for dates, UUID format for point_ids). Error handling is not visible in the provided code. No tool annotations (readOnlyHint, destructiveHint) despite all being read-only operations.
Execute a GraphQL query against the Tacit building digital twin API. Compose any query using the Brick-compliant schema. Supports nested fields, filtering by Brick class, supply chain traversal (upstream/downstream), and recursive location hierarchy. Use tacit_list_sites first to get a valid site ID, then construct queries freely.
List documents and files associated with a site. Returns metadata for files uploaded to a site: spec sheets, maintenance documents, BIM source files, 3D models, and knowledge graph data. Useful for answering questions like "What documentation exists for this building?" or "Are there spec sheets for this equipment?"
List all building sites the current API key has access to. Each site represents a physical location (building, campus, warehouse) managed in Tacit. Sites are the top-level container. You need a site ID to query buildings, equipment, points, zones, and systems. Returns: Array of sites with id, name, address, city, country, timezone. Use this tool first to discover available sites before querying building data.
Query historical or live sensor data for one or more points. Points are identified by their timeseriesId (UUID). Use tacit_graphql first to find points and their timeseriesId values.
Output schemas not documented. LLMs cannot see what fields to expect from tool responses (e.g., tacit_list_sites returns 'id, name, address, city, country, timezone' but this is only in the description, not in a structured schema). This forces LLMs to guess field names for downstream tool calls.
Parameter format constraints missing. 'point_ids' accepts 'comma-separated timeseriesId UUIDs' but no regex or format constraint is visible. 'start' and 'end' accept ISO 8601 or relative strings ('-1h', '-24h', '-7d') but format is not formally specified. LLMs may pass invalid formats.
No tool annotations despite all tools being read-only. MCP spec supports readOnlyHint to signal safe tools. This helps agents reason about side effects and retry safety.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | <=2025-11-25 | v2 |
Error handling not visible in source. No evidence of actionable error messages, error classification (retryable vs fatal), or recovery guidance. If API calls fail, LLMs receive raw errors with no guidance on next steps.
tacit_graphql accepts free-form GraphQL query strings. No validation visible. LLMs can pass malformed queries, and error messages from the GraphQL endpoint may not guide recovery. Consider wrapping common queries or validating schema compliance.