GitHub announced on June 17, 2026 that Agent Finder is now available in GitHub Copilot. In the official changelog, GitHub says Agent Finder can discover the right MCP servers, skills, canvases, agents, and tools for a task instead of forcing users to hand-wire them into every session.
That is a meaningful workflow update for QA engineers. Test automation work often needs different capabilities at different moments: browser debugging, log analysis, API inspection, pull-request review, or spreadsheet-style result analysis. When every agent session carries too much setup or too many tools, context gets noisy and governance gets harder.
What changed on June 17, 2026
- GitHub says Agent Finder now lets Copilot discover relevant AI resources on demand from a registry you choose.
- The changelog says results are ranked, so teams can decide which capability to install instead of connecting everything by default.
- GitHub also says Agent Finder does not auto-install tools silently, which keeps human approval in the loop.
- Microsoft published the open Agentic Resource Discovery (ARD) specification the same day, with GitHub named as a collaborator.
GitHub’s announcement also makes the governance angle explicit. Enterprises can scope discovery to a curated public catalog or a private internal registry, and Agent Finder only surfaces resources the organization permits.
Why this matters for QA engineers
For QA teams, the best agent workflows are rarely one-size-fits-all. A flaky Playwright failure may need browser and log tooling, while an API regression may need schema checks, contract-test helpers, or a custom review skill. Agent Finder matters because it can reduce setup friction without removing control.
- Less prompt plumbing: testers spend less time describing which tool stack an agent should use before useful work starts.
- Cleaner sessions: loading only task-relevant resources should reduce wasted context and improve auditability.
- Safer rollout: private registries and managed settings create a better path for approved internal QA skills and MCP servers.
- Better specialization: teams can keep separate capabilities for UI triage, CI failure review, API checks, and release evidence instead of building one bloated general-purpose agent.
What QA teams should test first
- Discovery quality: give Agent Finder a few real prompts such as flaky test triage, PR review, and API contract validation and see whether the ranked matches are actually useful.
- Approval controls: verify that suggested resources still require intentional installation or enablement and are not silently attached to sessions.
- Registry governance: if your team uses internal skills or MCP servers, confirm that only approved resources appear to testers.
- Task boundaries: compare whether narrower, task-specific tool loading leads to better outputs than broad always-on agent setups.
Why this matters for QA engineers
This is not a headline about raw model quality. It is a workflow and control update. For QA engineers evaluating AI agents in real testing work, that is often the more important change. If Agent Finder works as described, it should make AI-assisted testing setups easier to standardize, easier to govern, and less dependent on giant setup prompts.