TestRail MCP
Bidirectional TestRail connector. Pulls existing test cases and execution history; pushes EVAN-generated tests and run results back as cases, runs, and milestone reports. What flows in
·Test suite + section structure
·Existing test cases (steps, expected results, custom fields)
·Run history — pass/fail/blocked, durations, defect links
·Milestone definitions
What flows out
·New cases generated by EVAN (mapped to the right suite + section) ·Execution results from EVAN test runs ·Defects linked when SARA findings have a TestRail-tracked equivalent ·Milestone reports synthesized by ARIA Why bidirectional matters
→Many teams have years of TestRail history. The connector preserves that history rather than replacing it. EVAN-generated cases are added alongside hand-authored ones, with metadata tagging the source so audit + traceability survive the migration. Authentication
API key + email auth (TestRail's standard). Keys are stored encrypted, scoped per project.
Custom field mapping
TestRail allows arbitrary custom fields per project. The connector auto-discovers fields on first use and maps:
·Severity / priority
·Component / module
·Automation status (Manual / Automated / Not Required)
·Test type (Functional / Performance / Security)
Mappings are stored per-project to handle multi-project orgs.
Idempotency
When EVAN regenerates tests for a feature, the connector matches against existing cases by name + section + custom-field signature, not by ID. This means: ·Re-runs don't create duplicates
·Manual edits to test names will cause re-creation (intentional — engineers can use this as a "force reset" mechanism)
·A name match is treated as an update; the original case ID is preserved
Run management
·Create a new run per CI execution (default)
·Append to an existing open run (for long-running release campaigns)
·Update a milestone with aggregate results
The choice is configured per-project.
Vendor
TestRail (Idera). Server: mcp://testrail.io/mcp.
Limitations
→TestRail's API has stricter rate limits than Jira. Bulk imports are throttled aggressively; a 1000-case import takes ~10 minutes. The connector surfaces progress so the engineer knows it's not stuck.