Week of June 22, 2026
A lighter week on the surface, but the MCP server underneath caught up fast: search and read tools shipped alongside a first write path, verified end-to-end with a production smoke test. Feedback also picked up a privacy-preserving Recent view tagged with Capability Insights. Highlight: outcomet-mcp ships its first concrete tools, including direct feedback submission from a connected agent.
In this edition
Outcomet's MCP server goes live with real tools
The outcomet-mcp service earned its keep this week. It shipped its first concrete tools, search_feedback, get_feedback, search_capabilities, and get_capability, so a connected agent can actually look something up instead of just proving it can authenticate. A feedback theme search tool joined them, and permission enforcement was tightened so each tool only returns what its scope allows.
Read moreRead less
The bigger move was the first write path: add_feedback lets an authorized agent file a piece of feedback straight into the Outcomet inbox, the same capability that shows up on the product side as authorized MCP clients being able to submit feedback directly into the inbox. It's gated behind an explicit feedback:create scope and a service-level flag, so it's opt-in, not automatic. Alongside it, the OAuth client allowlist got tightened, the hosted demo's OAuth callback got fixed, and a production smoke test confirmed the whole path end-to-end.
- Concrete safe-read tools: search_feedback, get_feedback, search_capabilities, get_capability
- Feedback theme search tool for locating themes directly from a connected agent
- Tightened permission enforcement so each tool respects its configured scope
- First direct write tool, add_feedback, gated behind a feedback:create scope and explicit flag
- Authorized MCP clients can submit feedback straight into the Outcomet inbox
- OAuth client allowlist hardening and a fixed hosted demo OAuth callback, verified with a production smoke test
Feedback surfaces what it's about
A privacy-preserving Recent Feedback view now surfaces condensed summaries instead of raw text, so you can scan what's coming in without exposing what a user actually wrote. You can filter those summaries by timeframe and relevance to zero in on what's actionable, and feedback now carries Capability Insights tags so you can see which parts of the product are actually being talked about.
Read moreRead less
One access fix rounded out the week: the consent approval page now shows a clear, actionable error when it can't detect your signed-in account or workspace context, instead of just sitting there looking stuck.
- Privacy-preserving Recent Feedback view with condensed summaries instead of raw text
- Filter Recent Feedback Summaries by timeframe and relevance
- Feedback tagged with Capability Insights to show which product areas are referenced
- Consent approval page shows a clear, actionable error when account or workspace context can't be detected
What this means for you this week
- A connected MCP agent, Claude Code, Antigravity, or another client, can now search feedback and capabilities and file new feedback directly.
- Scan incoming feedback faster with a privacy-preserving Recent view tagged by capability.
- Approvals fail loudly instead of silently when your account or workspace context can't be detected.
If you want a walkthrough of connecting an agent through MCP, or the new Recent Feedback view, reply here and I'll set up a quick demo.