Your users report bugs and requests through the widget. AI triage categorizes and prioritizes them. This MCP server lets your coding agent inspect and operate on feedback without copy-pasting anything. Production release stays in the CLI because it needs the deployed Git revision.

User reports a bug  →  AI triage  →  Agent picks it up  →  Commit + ready  →  Deploy + release
     (widget)          (server)      (this MCP server)       (agent + MCP)          (CLI)

Install

No global install needed. Put this in your MCP client's user-level configuration to use the server across projects:

{
  "mcpServers": {
    "makethisbetter": {
      "command": "npx",
      "args": ["-y", "@makethisbetter/mcp@latest"]
    }
  }
}

Setup

1. Authenticate

If you have the CLI installed, just log in — it creates the config for you:

makethisbetter login

Or create the config manually:

mkdir -p ~/.makethisbetter
cat > ~/.makethisbetter/config.json << 'EOF'
{
  "api_token": "YOUR_API_TOKEN",
  "api_url": "https://makethisbetter.dev/api/v1"
}
EOF

Get your API token from makethisbetter.dev > Settings > API Tokens.

If you belong to more than one account, add "account_id": "YOUR_ACCOUNT_ID" — without it the server picks your first account, and projects in the other one come back as "not found". makethisbetter login --account-id YOUR_ACCOUNT_ID writes this field into the same file.

For self-hosted instances, point api_url to your own server (e.g. https://feedback.yourapp.com/api/v1).

2. Add to your editor

Run once to add the server at user scope:

claude mcp add --scope user makethisbetter -- npx -y @makethisbetter/mcp@latest

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "makethisbetter": {
      "command": "npx",
      "args": ["-y", "@makethisbetter/mcp@latest"]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "makethisbetter": {
      "command": "npx",
      "args": ["-y", "@makethisbetter/mcp@latest"]
    }
  }
}

Run MCP: Open User Configuration from the Command Palette, then add:

{
  "servers": {
    "makethisbetter": {
      "command": "npx",
      "args": ["-y", "@makethisbetter/mcp@latest"]
    }
  }
}

{
  "mcpServers": {
    "makethisbetter": {
      "command": "node",
      "args": ["/path/to/makethisbetter-mcp/dist/index.js"]
    }
  }
}

What your agent can do

Ten tools cover the Feedback lifecycle up to production release.

| Tool | What it does | |------|-------------| | list | Browse one Project's active or archived Feedback, filterable by status, priority, or an AI-selected label. archived: true cannot be combined with status | | detail | Read-only deep dive for active or archived Feedback. Returns server-rendered markdown for reading plus structuredContent with every raw field — selector, console errors, AI triage analysis, the reporter's annotations and breadcrumbs, the ai_clarification_messages exchange, the viewport (screen_width / screen_height / reporter_language), and ai_triage_status / ai_triage_error so a failed triage is not mistaken for one that never ran | | pick | Claim a feedback item. Sets it to in_progress and returns the same markdown + structuredContent context in one call | | respond | Send final user-confirmed text as a One-way Reporter Notice, queue delivery, and close received or in-progress Feedback as responded; the Agent must never generate and send the body autonomously | | archive | Hide one Unclaimed Feedback from active views without changing its workflow status | | restore | Return one archived Feedback to active views | | ready | Save the factual implementation summary and move committed work to pending_release; the calling Agent must first verify the exact Feedback trailer in reachable Git history | | decline | Close as not_planned without notifying the Reporter | | duplicate | Close as duplicate with a same-Project canonical_feedback_id | | reopen | Return closed Feedback to received; available to Account Owners/Admins, Active Pro Members, and assigned Team Members |

The ready tool does not inspect Git. The Agent must verify the exact Feedback: <handle/FB-n> trailer before calling it. release remains CLI-only because it scans the deployed revision and rejects shallow history.

respond, archive, and restore are retry-safe. A successful respond means its delivery record is persisted and queued, not that delivery has completed. Account Owners/Admins and Active Pro Members may perform these operations across the Account; Team Members may do so only for authorised Projects.

Four more tools manage the projects (feedback boards) themselves:

| Tool | What it does | |------|-------------| | project_list | Browse the projects in your account — id, name, domain, feedback visibility, feedback count | | project_show | Read full detail for one project, including its widget API key and board URL. Account Owners/Admins and Active Pro Members also receive the identity-verification signing secret | | project_create | Create a new project. name, handle and domain are all required — the widget only runs on the project's own domain. Available to Account Owners/Admins and Active Pro Members. Returns the widget API key and signing secret | | project_update | Update an existing project's name, domain, or ai_context. Available to Account Owners/Admins and Active Pro Members; the handle is immutable |

How it looks in practice

Here's what actually happens when your agent has this MCP server connected:

You:     "Any new feedback?"

Agent:   Let me check.
         → list(project_handle: "acme", status: "received")
         3 items. acme/FB-42 is a bug — "Login button doesn't work on Safari,
         rage-clicked 5 times." Looks real.

You:     "Fix it."

Agent:   Picking it up.
         → pick(feedback_id: "acme/FB-42")
         Got the full context: Safari 18.5, and the DOM snapshot puts the
         cookie banner on top of the login button. Console error:
         "Cannot read property 'submit' of null."

         The cookie banner has z-index: 9999 and covers the login
         button on viewports under 768px. I'll fix the z-index
         stacking and add a responsive margin.

         *makes the code changes*

         Fixed and tested. I will commit with `Feedback: acme/FB-42`, verify
         that trailer, then call ready with the factual implementation summary.

         → ready(feedback_id: "acme/FB-42",
                 resolution_summary: "Fixed the mobile cookie banner stacking order.")

         The production deployment agent releases it only after deployment.

No dashboard tab-switching. No copy-pasting bug descriptions. Feedback goes in, fix comes out.

Development

git clone https://github.com/makethisbetter/mcp.git
cd mcp
npm install
npm run build
npm test

# Run locally
MAKETHISBETTER_CONFIG=./config.json node dist/index.js

Related

| Package | What | |---------|------| | Make This Better | The platform — dashboard, AI triage, feedback board | | Widget SDK | Collect feedback from your website | | CLI | Manage feedback from the terminal | | Skills | Claude Code slash commands |

License

MIT