Ask your coding agent what customers want built.
Murmur ships a first-party MCP server. Point your agent at /api/mcp and it can rank requests, read the customer quotes, file new ones and close the loop when you ship.
› what should we build next? use murmur
● murmur.get_insights()
⎿ 4 themes · 12 open requests with revenue behind them
● murmur.search_posts(sort: "score", limit: 3)
⎿ 3 ideas
Here’s what customers are asking for, ranked by votes and revenue:
- 1.Google Sheets as a data source31 voters · $2.4k MRR
- 2.Dark mode for dashboards24 voters · $1.1k MRR
- 3.Weekly email digest17 voters · $640 MRR
#1 is blocking four paying teams. Want me to plan it and open a branch?
› yes — and tell the voters it’s underway
● murmur.update_post_status(status: "in_progress")
⎿ Status updated · voters notified
From “what next?” to “it’s live” without leaving your agent
Your coding agent already knows your codebase. Murmur tells it what your customers asked for — and lets it report back when the work is done.
- 01
Ask
“What should we build next?” — from the terminal or editor you’re already in.
get_insights - 02
Get a ranked answer
Your agent reads the board: votes, revenue, quotes and links to each request.
search_postsget_post - 03
Build it
Plan and ship with full customer context, then move the request to in progress.
update_post_status - 04
Close the loop
Mark it complete and Murmur notifies everyone who voted. Draft the changelog while you’re there.
update_post_statuslist_changelog
Connected in about two minutes
- 01
Create an API key
In your workspace, open Settings → Developers and create a key, or use Settings → MCP, which has the setup ready to copy. Keys are shown once and stored hashed.
- 02
Add Murmur to your agent
Run one command in Claude Code, or paste the JSON into Cursor or any client that supports remote MCP servers with headers. Replace
YOUR-APPwith your Murmur URL. - 03
Ask what to build
The Murmur tools appear in your agent. Ask in plain English — it picks the tools, and links every answer back to the request.
- Endpoint
- https://YOUR-APP/api/mcp
- Transport
- Streamable HTTP (JSON-RPC 2.0), stateless
- Auth
- Authorization: Bearer mur_live_…
- Keys
- Settings → Developers · Settings → MCP
- Plans
- Solo & Team
claude mcp add --transport http murmur \
https://YOUR-APP/api/mcp \
--header "Authorization: Bearer mur_live_…"Using a client that only speaks stdio? Wrap the endpoint with a remote-MCP bridge — the server follows the Streamable HTTP spec.
Fifteen tools, one workspace
Read tools look things up; write tools change your board and are recorded in the audit log with the actor “MCP”. Your agent sees the descriptions below.
Every board in the workspace, public and private, with its slug and number of ideas.
—
Search ideas by text, status and board; sort by trending, top, new, mrr, score or updated.
query?, status?, board?, sort?, limit?
One idea in full: votes, MRR, the companies behind it, AI summary and recent comments.
id
File a new idea on a board, optionally attributed to (and voted by) a customer.
board, title, body?, author_email?, author_name?
Move an idea to a new status. Voters are notified; an optional message is posted publicly.
id, status, message?
Reply publicly, or leave a team-only internal note customers never see.
id, body, internal?
Record a customer’s vote with their company and how much it matters. Idempotent per customer.
id, email, name?, company?, priority?
Run Autopilot on raw text — call notes, a support thread, a review — and queue what it finds for review.
text, source?, customer_name?, customer_email?, company_name?
The roadmap grouped by status, with votes, MRR and ETA.
—
Recently published changelog entries, newest first.
limit?
The last 90 days: recurring themes and the unshipped requests with the most revenue behind them.
—
Ideas you approved for Claude to build (Fix with Claude), with votes, MRR and progress.
include_finished?
The brief for an approved fix: the task, your instructions, the demand and the customer quotes, plus the rules (new branch, pull request).
id
Approve an idea for Claude to build. Only called after you say yes in the chat.
idea_id, note?
Report progress: working, pull request open, done or blocked. Done moves the idea to Complete and emails the voters.
id, status, note?, pr_url?, branch?
Things to ask your agent
A few prompts that fit a normal build day. The agent chooses the tools; you just ask.
Predictable by design.
An agent with write access to your customer feedback should be boring. Here’s exactly what a key can do.
Workspace keys
One key per agent or teammate, stored hashed, with last-used tracking. Revoke one without touching the rest.
Writes are logged
New posts, comments, votes and captures made over MCP are recorded with the actor “MCP”, with undo where possible.
Full workspace scope
A key acts for your team, so it can read private boards and internal notes. Only give keys to agents you trust.
Clear errors
Bad input comes back as a readable tool error, so agents correct themselves instead of guessing.
Keys are hashed at rest and shown once at creation.
- Any client that supports remote MCP servers over Streamable HTTP with a bearer header — including Claude Code, Cursor and agents built with an MCP SDK. For stdio-only clients, use a remote-MCP bridge.
- Solo and Team. Every new workspace can try it during the 14-day Team trial, no card required.
- Yes — an API key acts for your workspace team. Create a separate key per agent so you can revoke one at any time from Settings → Developers.
- Posts, comments, votes and captures it creates are in the audit log with an undo where possible. Status changes can be moved back like any other; revoke the key if you need to stop it entirely.
- The MCP server returns data only to the client you connected — your agent, under your account with that provider. Murmur doesn’t forward it anywhere else.
Give your agent the customer’s view.
Create a key, run one command, and ask what to build next. Included on Solo and Team — try it free for 14 days.