feedback-magic-mcp

A Model Context Protocol server for Feedback Magic. Point your AI editor (Claude Desktop, Cursor, …) at it and an agent can, without leaving the editor:

• see your workspace's projects and widgets,
• create a widget and get its publishable key — for surveys, pass a plain-English goal and AI designs the question set,
• generate a paste-ready embed snippet for your framework, and
• with a read-scope token: search feedback, read AI insights (weekly digest themes), and read survey results.

Tokens are provisioning + metadata only by default — they can create widgets and read project/widget metadata, but not feedback content. Reading content is a per-token opt-in: tick "Also allow reading feedback content & results" when creating the token.

Setup

1. In Feedback Magic, go to Settings → API & MCP and create a token. Copy the fm_sk_… value (shown once).
2. Add the server to your MCP client config, with the token in the env:

Claude Desktop (claude_desktop_config.json) / Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "feedback-magic": {
      "command": "npx",
      "args": ["-y", "feedback-magic-mcp"],
      "env": {
        "FEEDBACK_MAGIC_TOKEN": "fm_sk_your_token_here"
      }
    }
  }
}

Restart the client. The Feedback Magic tools will be available to the model.

Tools

| Tool | Args | What it does | | ---- | ---- | ------------ | | list_widget_types | — | Describe all six widget types (feedback, survey, annotate, changelog, board, testimonial). | | whoami | — | Show which workspace the token belongs to. | | list_projects | — | List the workspace's projects. | | list_widgets | project_id | List a project's widgets (incl. publishable keys). | | create_widget | project_id, type?, name?, goal? | Create a widget; returns its fm_pub_… key. goal (surveys) = AI-designed questions. | | get_embed_snippet | publishable_key, type?, framework? | Paste-ready embed (react / web-component / html / wordpress). | | search_feedback | project_id, query | read scope — search feedback titles/snippets. | | get_insights | project_id | read scope — latest AI digest + category mix. | | get_survey_results | widget_id | read scope — aggregated survey results. |

Typical flow: list_projects → create_widget → get_embed_snippet → paste into your app. (The get_embed_snippet output uses the feedback-magic npm package or the hosted <script> embed.)

Configuration

| Env var | Required | Default | | ------- | -------- | ------- | | FEEDBACK_MAGIC_TOKEN | yes | — | | FEEDBACK_MAGIC_URL | no | the hosted mcp endpoint | | FEEDBACK_MAGIC_APIKEY | no | the public Supabase anon key |

Security

The token is org-scoped and can be revoked anytime in Settings → API & MCP. By default it grants provisioning + metadata access to that one workspace — not feedback content, billing, or user management. The optional read scope (chosen at creation, shown as a badge in the token list) additionally allows reading feedback content, AI insights, and survey results — never billing or member management. Publishable keys it returns are safe to expose (they only permit submitting to that widget). Treat the fm_sk_… token itself as a secret (it's stored server-side only as an HMAC hash).

License

MIT