Loading ToolBench
Collibra Core REST MCP server for reading and writing asset metadata, attributes, and relations. Pairs with Power BI MCP servers to push semantic model details onto Collibra asset pages.
Server has 16 tools with consistent naming (verb_noun pattern: collibra_ping, collibra_find_assets, etc.) and all have descriptions. However, descriptions are often terse (10-50 chars) and lack actionable context for LLM selection. Parameter descriptions exist but are minimal. Output schemas are not documented, responses are raw API payloads without field-level guidance. Error handling is basic (RuntimeError with message). No tool annotations (readOnlyHint/destructiveHint). Write tools (create_asset, set_attribute, add_relation, import, request) lack confirmation/dry-run patterns despite irreversible consequences. Generic escape hatch (collibra_request) with IRREVERSIBLE risk is poorly constrained.
Add a relation (edge) between two assets.
Create an asset (e.g. a Measure/Column under a Data Set domain).
Search assets by name and/or asset type and/or domain. name_match: EXACT | START | END | ANYWHERE. Pass type_name (e.g. 'Report', 'Data Set', 'Column') and it is resolved to its id automatically.
Generic read-only GET against any /rest/2.0 path. Use for ad-hoc discovery. path: e.g. 'assetTypes', 'assets/<id>', 'jobs/<id>'. params_json: optional JSON object string of query params.
Get one asset, optionally with all its attributes and relations.
List all attributes (Description, Definition, custom fields, ...) of an asset.
Output schemas not documented. Tools return raw API responses (dicts) without field-level descriptions. LLMs cannot plan downstream calls or extract required data reliably.
Write tools (create_asset, set_attribute, add_relation, import, request) lack confirmation/dry-run patterns despite irreversible consequences. Agents can accidentally delete or corrupt data without safeguards.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Poll a bulk import job by ID. Returns status, progress, and any errors.
Bulk import assets and relations via the Collibra Import API. Returns a job ID; poll collibra_get_job() to track progress.
List asset types in the operating model (e.g. Report, Data Set, Table, Column, Measure).
List attribute types (Description, Definition, DAX expression, etc.).
List communities (top of the Collibra hierarchy).
List domains. Writes target a domain, so this finds where Edge put the PBI assets.
List relation types (source type, role/co-role, target type) - the lineage edges.
Verify connectivity and auth. Returns the current Collibra session/user and base URL.
Low-level escape hatch: make any HTTP request to the Collibra REST 2.0 API. Blocked on non-GET when COLLIBRA_READONLY is truthy.
Set an attribute (Description, Definition, custom field, etc.) on an asset. Creates or updates.
Generic escape hatch (collibra_request) accepts any HTTP method and path with minimal constraints. Marked IRREVERSIBLE but lacks input validation, rate limiting, or audit trail. High security risk.
Tool descriptions are terse (10-50 chars) and lack actionable context. E.g., 'List relation types' does not explain when to call it or what 'lineage edges' means. LLMs cannot reliably select the right tool.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes without parsing descriptions.