datanivra-mcp — local stdio MCP adapter (Universal Test Data API, Milestone 7)

A local Model Context Protocol server that lets an AI QA/coding agent discover and request pre-approved DataNivra test-data profiles through a small, least-privilege tool set. It is an interface to the ordinary Test Data API, not a bypass: every call goes through datanivra_sdk with the caller's own session and the control plane decides (AUTO_APPROVED, PENDING_HUMAN_APPROVAL, DENIED).

• Transport: stdio only (JSON-RPC 2.0, one message per line). There is no hosted or remote MCP endpoint (owner checkpoint D-05).
• Credentials: DATANIVRA_BASE_URL and DATANIVRA_TOKEN (a user session, or a CI workload session from datanivra login --workload-identity) from the environment. Tokens are never tool arguments and are never written to tool output.
• Tools (9): list_permitted_profiles, get_profile_schema, validate_test_data_request, plan_test_data_request, request_test_data, get_request_status, get_certification_evidence, get_lease_metadata, release_test_data_lease. There are no tools to approve requests, define or change profiles/policies, register sources, read secrets, run SQL, export data or change settings.
• Output: metadata only (ids, states, codes, counts, digests), size-bounded, marked as untrusted data.

Run: DATANIVRA_BASE_URL=https://api.example.test DATANIVRA_TOKEN=... datanivra-mcp. Install: pip install datanivra-mcp (PyPI). See examples/mcp/README.md.