QA.tech + Custom MCP Integrations
Custom MCP is the reverse direction of the MCP server: the QA.tech chat agent uses tools from services you connect via a remote MCP server, authenticated with OAuth, a bearer token, or no auth. Every tool gets a permission level: allow, needs approval, or blocked.
What teams connect
Examples that earn their keep
'%20gradientUnits='userSpaceOnUse'%3e%3cstop%20offset='0'%20stop-color='%230078d4'/%3e%3cstop%20offset='.16'%20stop-color='%231380da'/%3e%3cstop%20offset='.53'%20stop-color='%233c91e5'/%3e%3cstop%20offset='.82'%20stop-color='%23559cec'/%3e%3cstop%20offset='1'%20stop-color='%235ea0ef'/%3e%3c/linearGradient%3e%3c/defs%3e%3cpath%20fill='url(%23SVG6uMEBcPj)'%20d='M120.89%2028.445v69.262l-28.445%2023.324-44.09-16.07v15.93L23.395%2088.25l72.746%205.688V31.574ZM96.64%2031.93%2055.82%207.11v16.285L18.348%2034.418%207.109%2048.852v32.785l16.075%207.11V46.718Zm0%200'/%3e%3c/svg%3e)
Azure DevOps
Pull work items and acceptance criteria to ground test generation, then file bugs straight to Boards from chat.
Sentry
Cross-reference a failed test with the matching errors in your logs and attach the stack trace as evidence on the issue the agent files.
Notion
Let the agent read PRDs, specs and QA runbooks for context, and log test outcomes back into a Notion database.
Datadog Correlate a failing flow with APM traces to pinpoint the root cause.
Grafana Fetch backend log lines for the exact time window of a failed run.
Confluence Pull requirement pages and test plans as test context.
Asana Read tasks to ground tests in real requirements, and file bugs there.
ClickUp Keep bug reports in the tracker your team already lives in.
monday.com Turn board items into test cases, and report results back.
Snowflake Look up real records to seed a test with production-shaped data.
Postgres Verify the app wrote the right data end to end, not just the UI state.
Stripe Read customer and subscription state to set up or verify billing flows.
Zendesk Turn a support ticket into a reproduction test grounded in the report.
Intercom Pick up a reported bug from a conversation and reproduce it as a test.
…and any other MCP server, including your own internal tools.
What you can do
- Set a permission level per tool: allow, needs approval, or blocked
- Gate risky or write actions behind per-call approval
- Scope a server to selected projects instead of the whole organisation
Set it up
Connecting Custom MCP Integrations
- 01
Add your remote MCP server URL in QA.tech's Integrations settings.
- 02
Pick the auth method: OAuth, bearer token, or none.
- 03
Set a permission level per tool: allow, needs approval, or blocked.
Full reference in the QA.tech docs.
Where teams use this
Custom MCP Integrations in practice
The vivid version: a nightly run fails on checkout, the agent queries your Sentry MCP server for errors in that same window, finds the 500 behind it, and files an issue with the stack trace attached as evidence – so you get "failed because of this error", not just "failed". The same pattern works anywhere: pull acceptance criteria out of Azure DevOps or Notion before test generation writes a case, read real records from your database so CRM-heavy workflows are grounded in the actual system, or reach an internal admin panel we'll never build a native integration for.
FAQ
Common questions
- Which services can I connect?
- Any service with a remote MCP server – Azure DevOps, Sentry, Notion, Datadog, Confluence, Stripe, Zendesk, your own internal tools. If it speaks MCP, the agent can call it.
- When should I use Custom MCP vs a built-in integration?
- Built-in wins whenever it exists (Jira, Linear, TestRail, Xray, and the rest of this directory). Custom MCP covers everything else.
- Can I restrict what the agent may call?
- Yes, per tool: allow, needs approval, or blocked.
Related integrations
MCP Server
Next →Vercel Preview Protection
Your team moves fast. Can your testing keep up?
QA.tech agents test your product autonomously, so moving fast never means shipping broken. See how it works in a 30-minute demo.