Read your feedback from a coding agent

Feedstick speaks the Model Context Protocol, so Claude Code, Cursor, Codex, Gemini CLI — or any MCP client — can read your board directly. Ask “what are people asking for most?”, “summarize this week's bugs”, or “is anyone else reporting this?” — and have the answer drawn from real user feedback instead of guesswork.

MCP https://a.feedstick.app/mcp

Connect

Add the server to your agent, then sign in when your browser opens:

claude mcp add --transport http feedstick https://a.feedstick.app/mcp

{
  "mcpServers": {
    "feedstick": {
      "url": "https://a.feedstick.app/mcp"
    }
  }
}

codex mcp add feedstick --url https://a.feedstick.app/mcp
codex mcp login feedstick

gemini mcp add --transport http feedstick https://a.feedstick.app/mcp

Run /mcp inside Claude Code to sign in. Your browser opens to approve the connection.

Any other MCP client works the same way — point it at the URL above. Authentication is OAuth 2.1, so there is no API key to copy or paste anywhere: your client sends you to Feedstick, you pick which workspace to share and approve the permissions, and the client gets a token scoped to that one workspace. You can revoke it at any time from your dashboard.

Tools

list_boards

List the workspace's boards. A board is a saved filter over feedback (for example 'Bugs'), not a container — the same feedback can appear on several.

list_stacks

List stacks — groups of duplicate reports about the same thing — largest first. `count` is how many people reported it, so this is the fastest way to answer 'what do users want most?'.

search_feedback

Search the board by text, and/or filter by status and category. Returns stacked feedback with an excerpt of each item; use `get_feedback` for the full text of one. Done items are an archive and are EXCLUDED by default, exactly as they are on the board. When any are left out the result says so (`done_hidden`). Pass `include_done: true` to search the archive too, or `status: "done"` for only the archive.

get_board_feedback

Return the feedback matching one board's filter rules. Get board ids from `list_boards`.

get_feedback

Full detail for one feedback item: its complete text, its stack (every duplicate report of the same thing), and the team's comment thread.

list_tags

List the tags derived from submissions' `context` (for example `plan:pro`), with how many items carry each.

add_comment writes

Post a comment on a feedback item, attributed to you. The comment is visible to your team and notifies anyone watching. It is never emailed to the person who submitted the feedback.

merge_stacks writes

Merge one stack into another: every report in `source_stack_id` moves to `target_stack_id`, and the source stack is removed. Use this when two stacks are clearly about the same thing. The reports themselves are never deleted, so this can be undone with `move_feedback`.

move_feedback writes

Move one feedback item into a different stack, or — with `to_stack_id` omitted — split it out of its stack into its own. Use this to fix a report that was grouped with the wrong thing.

set_category writes

Change the category of one feedback item, or of a whole stack. Pass exactly one of `feedback_id` or `stack_id`. Recategorizing a single item that sits in a multi-report stack splits it out, since a stack holds one category.

set_status writes

Move a stack through the workflow. Setting it to `done` does NOT email the people who reported it — that is a deliberate action a human takes in Feedstick, never an agent.

Permissions

You choose what a connection can do when you approve it. Approve only the first and the agent gets a read-only connection; add the others and it can also comment and triage — merging duplicates, recategorizing, and moving items through your statuses.

feedstick:read

Read your feedback, boards, and comments

feedstick:comment

Post comments on your feedback as you

feedstick:triage

Reorganize your board — merge duplicates, recategorize, set status

What agents can and can't see

An agent reading your board never sees a submitter's email address. Feedback is attributed by the same anonymized handle the board shows you, and every response the server sends is scrubbed of email addresses on the way out — including addresses people typed into the feedback itself.

An agent also cannot email your customers. Comments posted through MCP are visible to your team and notify anyone watching the item, but the @-mention machinery that emails a submitter is not reachable from an MCP connection at all.

A note on untrusted feedback

Worth understanding before you connect an agent with other tools available to it. Feedback is submitted by members of the public — that is the point of it — which means the text your agent reads was written by people you don't control. Someone can submit feedback whose text is phrased as an instruction, hoping an AI agent reads it later and obeys.

Feedstick labels every piece of user-submitted content it hands to a model, and withholds the raw context blob unless a tool explicitly asks for it. But no server-side measure can fully separate an instruction from a feature request — they are the same sentence. Treat an agent connected to your board the way you'd treat one browsing the open web, and be deliberate about what other tools it can reach.

Getting feedback in

This page is about reading feedback. To send feedback to Feedstick, see the feedback API reference — one authenticated POST from a browser widget, your backend, or a script.