Run your feedback loop from your AI editor. No tab-switching.
Feedjolt ships a real MCP server — not a webhook in disguise. Connect Claude, Cursor, or any MCP client and ask your agent what users want, triage the inbox, and promote requests to the roadmap. The same actions you'd take in the dashboard, in plain language.
Your feedback lives in another tab.
You're shipping in your editor. The feedback is somewhere else — a dashboard you forgot to open. So requests pile up untriaged and the roadmap drifts from reality. The MCP server brings the inbox to where you already work, so reading and acting on feedback costs zero context-switches.
A real MCP server, not a toy integration.
Feedjolt speaks Model Context Protocol over Streamable HTTP. Your agent gets real tools, not a brittle scrape or a one-way webhook.
Connect Claude, Cursor, Windsurf, or any MCP client. One config block and the Feedjolt tools show up in your agent's tool picker.
Search and read feedback in natural language. "What are people asking for about login?" — your agent answers from real posts, votes, and comments.
Change statuses, tag posts, comment, and merge duplicates — all by asking. Clear the backlog without leaving your editor.
Move a winning request straight onto your roadmap by status, then notify everyone who asked. The loop closes from the command line.
Connect with a secure API token scoped to your own access. Your agent can only touch what you can — nothing more.
From editor to roadmap, in plain language.
Drop one MCP config block into Claude, Cursor, or your client of choice, with your Feedjolt API token. Restart and the tools appear.
"Show my top trending posts this week." "Summarize the login thread." Your agent reads real feedback and reports back.
Tell it to retag, change a status, merge duplicates, or promote a request to the roadmap. Done — without opening the dashboard.
The feedback loop, where your work happens.
Not a chatbot bolted onto a help doc. A real Model Context Protocol server your agent can call directly.
Read and act on feedback without a single tab-switch. The inbox comes to you, not the other way around.
The dashboard actions you rely on — statuses, tags, comments, merges, roadmap — all reachable by just asking.
Token auth scoped per user. The agent inherits exactly your permissions — no shared keys, no over-reach.
Frequently asked, plainly answered.
What is the Feedjolt MCP server?
It's a real Model Context Protocol server that exposes your Feedjolt feedback to AI agents. Connect it to Claude, Cursor, or any MCP client and your agent can read posts, triage the inbox, and promote requests to the roadmap — all in natural language.
Which MCP clients does it work with?
Any client that speaks MCP. It's documented for Claude Desktop, Cursor, and Windsurf, and any generic MCP client works too — the server speaks Streamable HTTP, so you POST JSON-RPC with your API token.
What can my agent actually do?
Read tools let it list and search posts, open threads, and view the roadmap and changelog. Triage tools let it change statuses, tag and untag posts, comment, and merge duplicates — the same actions you'd take in the dashboard, by asking.
Is it secure?
Yes. You connect with an API token scoped to your own access, passed as a bearer token. The agent can only touch what your account can — nothing is exposed beyond your existing permissions.
Can the agent create or delete boards and statuses?
No, by design. The MCP server doesn't create or delete posts, boards, statuses, or tags — schema changes stay in the dashboard. The agent triages and promotes within the structure you've already set up.
How is this different from a plain integration?
It's not a one-way webhook or a copy-paste export. It's a live MCP server your agent calls in real time, so it both reads the latest feedback and takes action on it — the full loop, from inside your editor.