Your feedback board,
readable by your agent.
FeatureWish ships a built-in Model Context Protocol server. Point Claude, Codex, Cursor, or any MCP client at your board and your AI assistant can pull the most-voted requests, read the comment threads behind them, check the roadmap, file new feedback, and close the loop when you ship — from wherever you already work.
https://app.featurewish.com/mcp Streamable HTTP · Authorization: Bearer fw_mcp_…Everything an agent needs to work the board.
Five read tools to understand your users, two write tools to act — nothing your assistant has to guess at, nothing it can reach beyond.
Read
list_boards List every feedback board in the workspace, with post counts.
list_posts Pull feedback posts — filter by board or status, sort by votes or recency.
get_post Read one request in full: body, status, vote count, and the whole comment thread.
get_roadmap The public roadmap as data — everything Planned, In progress, and Completed.
get_changelog Published changelog entries, so your agent knows what already shipped.
Write
create_post File a new feedback post — e.g. log a request that arrived by email or support chat.
update_post_status Move a request through the lifecycle. Status changes email the voters automatically.
Deliberately scoped. A token grants access to one workspace's feedback — never billing, team management, or another workspace. Revoke any token instantly from Settings → MCP.
Connected in two minutes.
Generate a token in Settings → MCP, then paste one snippet into the tool you already use. No SDK, no OAuth dance, no webhook wiring.
Claude Code
claude mcp add --transport http featurewish \
https://app.featurewish.com/mcp \
--header "Authorization: Bearer fw_mcp_YOUR_TOKEN"Cursor (.cursor/mcp.json)
{
"mcpServers": {
"featurewish": {
"url": "https://app.featurewish.com/mcp",
"headers": {
"Authorization": "Bearer fw_mcp_YOUR_TOKEN"
}
}
}
}Codex CLI (~/.codex/config.toml)
[mcp_servers.featurewish]
url = "https://app.featurewish.com/mcp"
bearer_token = "fw_mcp_YOUR_TOKEN"Claude Desktop & Claude.ai
Add a custom connector, paste the endpoint URL, and supply the token as a bearer Authorization header. Your assistant discovers the seven tools by itself the first time it connects.
Things you can just… say.
Once connected, your feedback board answers in plain language.
Your agent knows your code. Now it knows your users.
If you build with an AI agent, you've felt the asymmetry: your assistant can read every file in your repository, but it knows nothing about what your users are asking for. It can tell you how to build anything — and can't help you decide what to build, because the demand signal lives in a browser tab it can't see.
The MCP server closes that gap. Your feedback board — every request, every vote, every comment thread explaining why someone needs something — becomes context your agent can query mid-conversation. Planning a sprint inside Claude Code? Ask which open requests have the most votes and the least effort. Wondering if anyone actually wants the thing you're about to spend a week on? The board answers before you write a line.
And it works in both directions. When you ship, your agent marks the request completed — which emails every voter automatically. The loop that makes feedback boards compound (users see that asking works, so they keep asking) now runs from the same chat window you built the feature in.
We wrote a full guide on the workflows this unlocks for small product teams — FeatureWish for AI agents — including morning triage, spec-drafting from comment threads, and closing the loop without leaving your editor.
MCP server questions, answered.
What is the FeatureWish MCP server?
It’s a built-in Model Context Protocol endpoint for your workspace. MCP is the open standard AI assistants use to talk to external tools — point Claude, Codex, Cursor, or any MCP client at your FeatureWish board and your assistant can list requests, read comment threads, check the roadmap, file feedback, and update statuses on your behalf.
Which AI tools does it work with?
Anything that speaks MCP over HTTP with a bearer token: Claude Code, Claude Desktop, Claude.ai custom connectors, OpenAI’s Codex CLI, and Cursor all work today, and any other MCP-compatible client can connect the same way. There is no SDK to install — it’s one URL and one token.
Is the MCP server free?
It’s part of the Pro plan — $20/month flat, alongside five workspaces with ten boards each, the GitHub/Linear integrations, custom CSS, and white-labeling. The price never scales with usage: no per-call metering, no per-seat fees, no tracked-user math.
Is it safe? What can the agent actually touch?
Each token is scoped to exactly one workspace and can be revoked at any time from Settings → MCP. The agent sees your feedback data — boards, posts, comments, roadmap, changelog — and can create posts and change statuses. It cannot touch billing, team members, other workspaces, or anything outside the seven documented tools.
Do I need to write any code?
No. Generate a token in Settings → MCP, paste the endpoint URL and token into your AI tool’s connector settings, and start asking questions. The setup snippets on this page are copy-paste.
How is this different from a REST API?
A REST API is built for programs you write; MCP is built for agents you talk to. Your assistant discovers the tools by itself, understands their descriptions, and chains them — “find the top request, read its comments, draft a spec” is one sentence, not an integration project.
Give your agent the demand signal.
Start free, collect feedback, and flip on the MCP server whenever you're ready — $20/mo flat, never metered.
Create your board — it's free