Featurebase MCP server

Connect AI tools to your Featurebase workspace and manage feedback, support, and product data through natural language.

Written By Markus from Featurebase

Last updated 5 days ago

Overview

The Featurebase MCP server lets AI tools and assistants like Claude, ChatGPT, and Cursor securely connect to your Featurebase workspace.

Once connected, AI agents can read, create, and update data across Featurebase – no custom integration work required.

What is Featurebase MCP?

MCP (Model Context Protocol) is an open standard for connecting AI tools to external systems, so any MCP-compatible client can talk to Featurebase out of the box.

The MCP server is built directly on top of our public API, which means anything you can do with the Featurebase API, an AI agent can do for you in plain conversation:

  • Easy setup – Connect with OAuth in a few clicks and start managing your workspace through chat

  • Works with any MCP client – Claude, ChatGPT, Cursor, VS Code, Claude Code, and any other client that reads an mcpServers.json (Continue, Zed, Windsurf, etc.)

  • Full workspace access – Manage feedback, support conversations, users, changelogs, Help Center articles, and more

  • Secure by design – AI tools authenticate as your Featurebase user and inherit your existing permissions

Note: AI agents acting through the MCP have the same access you do, so only connect tools you trust.

Reader and Writer: two connectors, one safety boundary

The hosted MCP ships as two separate connectors – Reader and Writer – so mutation tools never share a chat with customer text that could trick an agent into running them. Pick whichever one fits the job.

  • Reader – Read-only access to your workspace: feedback, conversations, contacts, help center. Safe for analytics, reporting, and Q&A agents.

  • Writer – Create, edit, and delete admin-controlled resources (boards, tags, help-center articles, webhooks). Intentionally cannot read customer-supplied content.

Never use Writer in the same chat as Reader.

Writer is safe on its own – it can't read customer text, so there's nothing to inject instructions into. The risk only appears when you put it in the same chat as Reader. An attacker can hide instructions inside a feedback post, support reply, or comment that Reader pulls in; the agent then runs those instructions as real writes through Writer's tools – a high chance of compromising your Featurebase organization. Keep Writer in a separate Claude project / ChatGPT chat / Cursor workspace.

What you can do with each connector

Reader (read-only, customer content allowed)

  • Feedback & Roadmaps – Search and read feedback posts, comments, votes, voters

  • Support Inbox – Read conversations, tickets, and customer history

  • Help Center – Read articles and collections

  • Users & companies – Look up customer profiles, segments, and attributes

  • Surveys – Pull responses for analysis

  • Changelog – Read past releases and subscribers

Best for: weekly digests, sentiment analysis, "what are users complaining about", customer-history summaries before a reply, Q&A bots over your help center.

Writer (admin actions, no customer content)

  • Feedback boards – Create, rename, reorder, archive

  • Tags & custom fields – Create, rename, delete, bulk-organize

  • Help Center – Create, read, edit, restructure, and publish articles and collections

  • Changelog – Draft, edit, and publish releases

  • Webhooks & integrations – Register, rotate secrets, remove

  • Post statuses, ticket statuses, conversation tags – Configure

Best for: building out your information architecture, drafting changelogs from notes you paste in, restyling old help articles, bulk admin cleanup.

Tip: Workflows that need both – e.g. "find recurring questions in the inbox and draft new help articles" – are best done in two passes. Use Reader in one chat to extract and summarize, then paste the summary into a separate Writer chat to create the articles. For the full list of available actions per connector, see our API reference.

Getting started

Connect Featurebase with your AI tool:

  1. Before connecting Featurebase MCP, ensure you have:

    • An active Featurebase workspace (on the Professional plan)

    • An MCP-compatible AI client (Claude.ai, ChatGPT, Cursor, VS Code, Claude Code, etc.)

  2. Go to Featurebase Settings → MCP

  3. Pick Reader or Writer, then follow the install card for your AI tool (Claude, ChatGPT, Cursor, VS Code, Claude Code, or Other clients via mcpServers.json)

You can revoke MCP access at any time from your Featurebase account settings.

Note: If your AI tool doesn't yet support remote MCP servers, you can connect through the mcp-remote proxy package as a bridge.

On Claude Team or Enterprise plans? Custom connectors must be enabled at the organization level by an Owner or Primary Owner before individual members can install them. If you see Missing permissions or Failed to add connector when trying to install Featurebase, that's why.

The Owner needs to go to Organization settings → Connectors → Add, paste the Featurebase Reader (or Writer) URL, and click Add. Once it's enabled at the org level, individual members can connect from Customize → Connectors as normal. See Claude's documentation on custom connectors for the full flow.

Example use cases

Here are a few ways teams use the Featurebase MCP day-to-day. Each bullet is tagged with the connector it needs.

Support suite

  • Chat with your Help Center (Reader) – Ask questions across your full documentation and get instant answers grounded in your actual articles

  • Draft and update Help Center articles (Writer) – Write new articles, refresh outdated ones, or restyle old content to match your current voice – all through chat

  • Audit your docs for gaps (Reader + Writer in separate chats) – Have AI scan your Help Center against recent product changes in a Reader chat, then apply the rewrites in a Writer chat

  • Generate articles from support trends (Reader + Writer in separate chats) – Find recurring questions in your Inbox with Reader, paste the summary into a Writer chat to draft the articles

  • Triage and draft replies (Reader) – Pull conversation context from the Inbox and have AI draft personalized replies grounded in the customer's history

  • Surface conversations by customer or topic (Reader) – Ask things like "show me every open issue from Acme Corp" or "summarize all chats about the new billing flow this month from top paying customers" and get a grouped, themed overview pulled straight from the Inbox

Product suite

  • Analyze feedback trends (Reader) – Ask about themes, sentiment, or feature interest across all incoming feedback ("What are users complaining about most this month?")

  • Turn call transcripts into feedback posts (Writer) – Build a custom Claude Cowork automation that pulls insights from sales or support call transcripts and creates new posts. Keep this in a Writer-only chat – paste transcripts in directly rather than reading them through Reader in the same session.

  • Triage incoming feedback (Reader + Writer in separate chats) – Read and cluster posts in a Reader chat, then apply tags and merges in a Writer chat

  • Draft changelogs (Writer) – Generate changelog releases from notes you paste in, with proper categorization and release notes

  • Generate product reports (Reader) – Get weekly or monthly digests of feedback grouped by tag, status, or customer tier

  • Analyze survey responses (Reader) – Pull patterns and sentiment from open-ended survey answers across hundreds of responses

General

  • Summarize customer history (Reader) – Get a full summary of a customer's past conversations, feedback, and feature requests before you reply

  • Bulk user data management (Writer) – Update tags, attributes, or segments using lists you paste into the chat

FAQ