ShipLog

npx add-mcp 'https://shiplog.ca/mcp'

Your AI agent reads the feedback and writes the changelog

ShipLog runs a remote MCP server. Point Claude Code, Claude Desktop, Cursor or any MCP client at it with a project API key, and your agent can pull your top-voted feedback into a coding session, ask what to build next, mark an item shipped, and publish the changelog entry — without leaving the editor. Seven tools, no local process to run, and it's on every plan including the free one.

Connect your agent See the tools

Connect in two steps

1. Create a project API key

Your agent authenticates with a project API key from Settings → API keys. It's shown once, scoped to that one project, and revocable at any time. If you don't have an account yet, the free plan includes this — create one and you'll have a key in about a minute.

2. Point your client at it

For Claude Code, put the key in an environment variable and run this in your project:

export SHIPLOG_API_KEY=sl_live_YOUR_KEY

claude mcp add --transport http shiplog https://shiplog.ca/mcp \
  --header 'Authorization: Bearer ${SHIPLOG_API_KEY}'

The single quotes matter. They keep ${SHIPLOG_API_KEY} a literal in the stored config, so your key is read from the environment when the tool runs instead of being written to disk. Don't paste a raw key into a project's.mcp.json — it ends up in git history.

For Claude Desktop, Cursor, or anything else that takes a Streamable HTTP server:

{
  "mcpServers": {
    "shiplog": {
      "type": "http",
      "url": "https://shiplog.ca/mcp",
      "headers": {
        "Authorization": "Bearer ${SHIPLOG_API_KEY}"
      }
    }
  }
}

The endpoint is https://shiplog.ca/mcp and it speaks JSON-RPC over Streamable HTTP. It negotiates protocol versions 2025-06-18, 2025-03-26 and 2024-11-05, so an older client still connects. Opening that URL in a browser returns 401 — it's the server, not a page.

The tools your agent gets

All of them, on every plan. This list comes from the server's own tool definitions, so it's what you'll actually see when your client calls tools/list.

list_feedback

List the project's feedback items, sorted by votes (default) or date. Use this to see what users are asking for most.

search_feedback

Search feedback items by keyword in the title or description.

feedback_summary

Get an overview of the project's feedback: total count, counts by status, and the top-voted items. The best starting point for "what should I build next?".

create_feedback

Add a feedback item to the project's board.

update_feedback_status

Move a feedback item to a new status (e.g. mark it planned, in_progress or completed).

list_changelog

List the project's changelog entries, newest first, including drafts.

create_changelog_entry

Write a changelog entry (the captain's log). Drafts by default; set publish to true to publish it immediately.

There is nothing destructive in that list on purpose. An agent cannot delete feedback, cannot touch billing, and cannot create or revoke API keys. Every call is scoped to the single project its key belongs to.

What the loop looks like

This is the whole reason the MCP server exists. Four steps, all from your editor.

1. 1 Your agent asks what to build It calls feedback_summary and gets your board's top-voted items with their vote counts and statuses — not a wall of raw text to re-read every session.
2. 2 You ship it The part nobody can automate for you.
3. 3 The item is marked shipped update_feedback_status moves it to completed, so your board reflects reality without you opening it.
4. 4 The changelog entry gets written create_changelog_entry drafts it from what actually changed and publishes it to your public changelog, which has an RSS feed readers can follow.

We use it on this product

ShipLog's own changelog is written this way. The entries on our public changelog were drafted and published by an agent calling create_changelog_entry against this server, from the feedback on our public board — the same two pages any ShipLog project gets. Go read them; that's the output.

Included on every plan, free tier included

There is no AI add-on and no tier where the agent access lives. The free plan is limited to one project with 10 feedback items and 10 changelog entries, including drafts. API and MCP access are included. Captain is $9/mo per project and lifts both limits.

The incumbents charge for it — Canny lists MCP under Pro at $79/mo, Sleekplan under its paid plans, Upvoty under Hyper at $49/mo. Some newer tools do offer free MCP access, so the claim worth making isn't that ShipLog is the only free MCP server. It's that the complete loop — read the feedback, publish the changelog — is free here, and isn't on the tools most people are comparing against. See the comparison .

Frequently asked questions

Does the ShipLog MCP server work with Claude Code and Cursor?

Yes, and with any client that speaks MCP over Streamable HTTP. It's a remote server, so there's nothing to install and nothing to run locally — you point your client at https://shiplog.ca/mcp with a project API key. The server negotiates protocol versions 2025-06-18, 2025-03-26 and 2024-11-05, so older clients keep working.

Is MCP really included on the free plan?

Yes. Core features are on every plan, including the free one, which is capped at 10 feedback items and 10 changelog entries for one free project. The incumbents charge for it: Canny lists MCP under Pro ($79/mo), Sleekplan under its paid plans, Upvoty under Hyper ($49/mo). Other newcomers do offer free MCP access, so the honest claim isn't that ShipLog is the only one with a free MCP server — it's that the complete loop, reading feedback and publishing the changelog entry, is free here and isn't on the incumbents.

What can an AI agent actually change in my project?

It can read feedback, search it, add an item, change an item's status, list changelog entries and write one. That's the whole surface. There are no destructive tools: an agent cannot delete anything, cannot touch billing, and cannot create or revoke API keys. Every call is scoped to the one project its key belongs to, so a key can never read or write another project's data.

Does publishing a changelog entry email my users?

No. ShipLog sends exactly one kind of email — your magic login link. Publishing a changelog entry puts it on your public changelog page, which has an RSS feed readers can follow. Nobody is emailed, and voters are not notified that the thing they voted for shipped. Canny does send changelog emails on its paid plans; that's a real capability ShipLog doesn't have, and we'd rather say so than let you find out after switching.

How do I revoke a key if an agent misbehaves?

Settings → API keys, and it stops working immediately. Keys are stored hashed and show when they were last used, so you can tell which one an agent is on before you revoke it. Requests are rate limited per project. Feedback is end-user submitted text, and the server tells your agent to treat it as data rather than instructions — still worth reading a changelog entry your agent drafted from feedback before you publish it.

Give your agent a seat on deck

Free plan, one project, 10 feedback items and 10 changelog entries, no credit card. You'll have a key and a connected agent in about a minute.