MCP Server

MANAGANI includes a Model Context Protocol (MCP) server for Claude Desktop, Cursor, and other MCP-compatible clients.

Connect once, then let compatible AI tools work with sites, URLs, feedback, roadmap, changelog, and widget conversations.

Features

• Tools across sites, URLs, feedback, roadmap, changelog, and widget chat
• Shared site context — use the same site ID across product workflows
• Fast setup — connect most MCP clients in about a minute
• Automatic default site — tools work without specifying a site
• Three transports: stdio (default), SSE, and Streamable HTTP
• Token-based auth via Authorization: Bearer, access-token, or stdio env vars

Quick Start

# Run with stdio transport (default)
env 'ACCESS-TOKEN'=your-access-token API_BASE_URL=https://your-instance.com node apps/mcp/server.js

# Run with HTTP transport
node apps/mcp/server.js --transport http --port 3002

See Setup for Claude Desktop configuration and Tools for the full tool reference.

For Cursor (global User Rules, site .cursor/rules, and MCP server naming), see Cursor (IDE).

Tool Categories

Category	Tools	Description
URLs	6	Save, list, search, read, update, delete URLs
Sites	6	Create, read, update, delete, list sites, and get counts
Feedback	6	List, read, create, update, comment, and suggest replies
Roadmap	2	Read roadmap and move feedback to roadmap lists
Changelog	4	List, draft, update, and publish changelog entries
Widget Chat	3	List, read, and reply to widget conversations

Start with Read-Only Context

After connecting a client, ask it to retrieve a known site or feedback item before requesting changes. Confirm that the result belongs to the intended account and site. A default site is convenient, but an explicit site is safer when the same client works across several products.

Use narrow retrieval requests so the assistant receives the context needed for the current task. Read the most relevant records before broadening a search. For changes that affect customers, such as replies or published updates, tell the assistant which actions you want performed and review the proposed content.

Add recurring expectations to your agent instructions so future sessions use the same retrieval and approval workflow.

For repeatable team workflows, record the selected site and the expected retrieval steps in the client instructions.

For example, a request to summarize recent feedback should begin with the relevant site's feedback records and return links or identifiers that the team can inspect. Publishing a changelog is a separate action: verify the draft and release context before asking the connected client to make that update public.