generated: '2026-08-11'
method: derived
source: >-
  mcp/credo-ai-mcp.yml (tools transcribed from
  github.com/credo-ai/credoai-plugins) bound against
  openapi/credo-ai-governance-platform-swagger.json and
  openapi/credo-ai-audit-logs-shadow-ai-openapi.json
surfaces:
  openapi:
  - file: openapi/credo-ai-governance-platform-swagger.json
    spec: Swagger 2.0
    base: https://api.credo.ai/api/v2/{tenant}
    operations: 300
    gated: true
    note: >-
      The document itself is public and unauthenticated at
      https://api.credo.ai/swagger.json; every operation it describes returns 401
      without a tenant Bearer token.
  - file: openapi/credo-ai-audit-logs-shadow-ai-openapi.json
    spec: OpenAPI 3.0.0
    base: https://api.credo.ai/api/v2/{tenant}
    operations: 11
    gated: true
  graphql: null
  mcp:
    name: governance-hub
    transport: stdio
    package: '@credoai/governance-hub-mcp'
    gated: true
    note: >-
      npm returned 404 on 2026-08-11; tools/list could not be introspected, so no
      tool inputSchema is available and every binding below is by name and
      documented semantics, not by observed schema.
headline_finding: >-
  The MCP server and the REST API are NOT two projections of one core. The
  governance-hub MCP reads the CANONICAL, cross-tenant Credo AI Governance
  Intelligence catalog — the risk/control/policy-requirement taxonomy and the
  maturity framework Credo AI authors — while the v2 REST API operates on a
  SINGLE CUSTOMER TENANT's instantiation of that taxonomy. Every MCP tool is
  therefore mcp_only, and 300 of 311 REST operations are rest_only. An agent
  cannot reach the catalog over REST, and cannot write a use case over MCP.
crosswalk:
- tool: governance_query
  category: catalog-search
  rest: []
  binding: none
  confidence: high
  note: >-
    Nearest REST neighbours are CredoAIWeb.API.V2.PolicyControlController.index,
    CredoAIWeb.API.V2.RiskScenarioController.index and
    CredoAIWeb.API.V2.RiskTypeController.index, but those list the controls and
    risk scenarios already attached to the caller's tenant. They are not a
    semantic search over the master catalog and cannot substitute.
- tool: get_entities
  category: catalog-read
  rest: []
  binding: none
  confidence: high
  note: >-
    Tenant-scoped analogues exist
    (CredoAIWeb.API.V2.PolicyControlController.show,
    CredoAIWeb.API.V2.RiskScenarioController.show,
    CredoAIWeb.API.V2.PolicyPackController.show) but they resolve tenant IDs, not
    catalog result_ids returned by governance_query.
- tool: get_catalog_overview
  category: capability-probe
  rest: []
  binding: none
  confidence: high
  note: >-
    A connection/feature probe with no REST equivalent. The closest REST call is
    CredoAIWeb.API.V2.TenantInfoController.show, which describes the tenant rather
    than the catalog.
- tool: get_maturity_framework
  category: framework
  rest: []
  binding: none
  confidence: high
  note: >-
    Pro-entitled and metered. The maturity framework is not exposed anywhere in
    either OpenAPI document.
- tool: get_benchmarks
  category: benchmarks
  rest: []
  binding: none
  confidence: high
  note: Pro-entitled and metered. No REST equivalent in either OpenAPI document.
mcp_only:
- tool: governance_query
  reason: >-
    Semantic/vector search over the cross-tenant Governance Intelligence catalog.
    No public REST operation exposes the catalog.
- tool: get_entities
  reason: Resolves catalog record IDs, which the tenant REST API does not address.
- tool: get_catalog_overview
  reason: MCP-session capability probe with no REST analogue.
- tool: get_maturity_framework
  reason: Entitlement-gated framework document; absent from both OpenAPI contracts.
- tool: get_benchmarks
  reason: Entitlement-gated benchmark corpus; absent from both OpenAPI contracts.
rest_only_summary:
  total_rest_operations: 311
  reachable_via_mcp: 0
  note: >-
    Listed by tag rather than by 311 rows. Not one write path in the platform is
    agent-reachable through the published MCP server.
  by_tag:
  - tag: use_cases
    operations: 24
  - tag: vendors
    operations: 23
  - tag: models
    operations: 27
  - tag: use_case_controls
    operations: 14
  - tag: policy_packs
    operations: 13
  - tag: questionnaires
    operations: 13
  - tag: risks
    operations: 12
  - tag: fields
    operations: 12
  - tag: triggers_and_actions
    operations: 12
  - tag: use_case_questionnaires
    operations: 12
  - tag: use_case_reviews
    operations: 12
  - tag: policy_controls
    operations: 11
  - tag: entity_controls
    operations: 10
  - tag: use_case_reports
    operations: 8
  - tag: tasks
    operations: 7
  - tag: teams
    operations: 7
  - tag: comment_threads
    operations: 7
  - tag: integrations
    operations: 7
  - tag: user
    operations: 11
  - tag: users
    operations: 4
  - tag: alerts
    operations: 5
  - tag: workflow_stages
    operations: 5
  - tag: use_case_scheduled_reviews
    operations: 5
  - tag: use_case_policy_packs
    operations: 5
  - tag: use_case_custom_fields
    operations: 5
  - tag: use_case_risk_scenarios
    operations: 5
  - tag: entity_risk_scenarios
    operations: 5
  - tag: sources
    operations: 4
  - tag: entity_policy_packs
    operations: 4
  - tag: entity_risk_scenario_controls
    operations: 4
  - tag: other
    operations: 3
  - tag: 'Shadow AI (OpenAPI 3.0)'
    operations: 7
  - tag: 'Audit Logs (OpenAPI 3.0)'
    operations: 4
  - tag: remaining single-operation tags
    operations: 6
  - tag: webhooks
    operations: 0
    note: >-
      Declared as a tag in the Swagger document with a description ("Operations
      related to webhooks") but carrying ZERO operations. A declared-but-empty
      tag, not an event surface.
coverage:
  mcp_tools: 5
  mcp_tools_bound_to_rest: 0
  mcp_only: 5
  rest_operations: 311
  rest_only: 311
  overlap: 0
  overlap_percent: 0.0