MCP Server

Connect Claude — or any MCP-compatible AI assistant — to ProductBridge with a single bearer token. Let Claude read your feedback, search semantically, create posts, and update the roadmap from any conversation.

What Is the MCP Server?

ProductBridge ships an MCP (Model Context Protocol) Server that exposes your workspace to AI assistants. Once connected, Claude can:

Search and read feedback, roadmap items, and changelog entries
Find semantically similar feedback ("anything else that sounds like this?")
Create new feedback posts, roadmap items, and changelog entries
Look up boards, statuses, tags, categories, and user segments
Filter feedback by user segment ("what are Enterprise customers asking for?")
Upvote feedback on behalf of a user

The connection is bearer-token authenticated and scoped to your organization — Claude only ever sees your workspace's data.

MCP is an open protocol from Anthropic. While these docs focus on Claude Desktop and Claude.ai, the same connection works with any MCP-compatible client.

How It Works

Claude calls a tool, the MCP server verifies your bearer token, scopes the call to your organization, runs it against your data, and streams the result back. No data leaves your workspace's boundary.

Connect Claude in Three Steps

Generate your connection config

In the ProductBridge admin, go to Settings → Connect → MCP Server and click Generate MCP Config.

Optionally name the connection (the Credential name field) so you can identify it later — useful if you create multiple connections for different machines or pipelines.

ProductBridge generates:

A Server URL to connect to
A Bearer token (shown with a reveal/hide toggle — copy it from the field)
A Key prefix — a short identifier for the token used in audit displays
Ready-to-paste config snippets for Claude Desktop and Claude Code

Treat the bearer token as a production secret. If you suspect it's exposed, click Regenerate to issue a fresh one — the old key becomes invalid immediately.

Paste into your Claude client

Open your Claude Desktop config file:

macOS / Linux: ~/.claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Paste the config snippet ProductBridge generated:

{
  "mcpServers": {
    "productbridge": {
      "type": "http",
      "url": "https://api.productbridge.io/mcp",
      "headers": {
        "Authorization": "Bearer pb_your_token_here"
      }
    }
  }
}

Restart Claude Desktop for the change to take effect.

Try it out

In Claude, try a prompt like:

"Use ProductBridge to list the top 10 feedback posts from Enterprise customers."

Claude will call the appropriate MCP tools — you'll see the tool calls appear inline — and respond grounded in your real data.

What Claude Can Do

The MCP Server exposes 27 tools across four areas. Claude picks the right ones automatically based on your prompt.

Feedback

Tool	What It Does
`get_feedback_list`	List feedback posts with filters (status, category, search, pagination)
`search_similar_feedback`	Semantic vector search — "find anything that sounds like this"
`get_feedback_by_segment`	List feedback posts from users in a specific segment
`create_feedback`	Create a new feedback post
`update_feedback`	Update title, description, board, status, category, or tags
`upvote_feedback`	Toggle an upvote on a post

Roadmap

Tool	What It Does
`get_roadmap`	List roadmap items grouped by status
`create_roadmap_item`	Create a new roadmap item with title, description, ETA, effort, impact

Changelog

Tool	What It Does
`get_changelog`	List changelog posts with status, type, and label filters
`create_changelog`	Create a new changelog entry (draft or published)

Configuration & Lookups

Tool	What It Does
`list_feedback_boards`	List all feedback boards in your workspace
`list_roadmap_boards`	List all roadmap boards
`list_feedback_statuses` / `list_roadmap_statuses`	List available statuses
`list_feedback_categories` / `list_roadmap_categories`	List available categories
`list_feedback_tags` / `list_roadmap_tags`	List available tags
`list_changelog_status_codes` / `list_changelog_types` / `list_changelog_labels`	List changelog metadata
`list_user_segments`	List user segments — use these IDs with `get_feedback_by_segment`

Claude reads the tool descriptions, decides which ones it needs, and composes them. You don't need to memorize the names — just ask in plain English.

Example Conversations

"I just got an email from a customer asking for dark mode. Can you check ProductBridge for similar requests and tell me how many votes the top match has?"

Claude calls search_similar_feedback with your query, returns the top matches, and reports the vote counts.

"Pull the top 10 highest-voted open feedback from our Enterprise Customers segment and group them by category."

Claude calls list_user_segments to find the Enterprise segment, then get_feedback_by_segment to fetch the posts, and groups the response itself.

"Here's a Slack thread about the new SSO update we just shipped: [paste]. Draft a changelog entry as a feature, label it Auth, save it as a draft."

Claude calls list_changelog_types and list_changelog_labels to find the right values, then create_changelog with the drafted content.

"Add a roadmap item titled 'Workspace-level audit logs' on the Security board, effort 5, impact 8, ETA end of next quarter."

Claude calls list_roadmap_boards to find the Security board, then create_roadmap_item with the specified fields.

Connection Management

Regenerate Your Token

If your token is exposed or you want to refresh it:

Go to Settings → Connect → MCP Server
Click Regenerate

The old token is invalidated immediately. You'll need to update your Claude config with the new one.

Credential Name & Key Prefix

Each connection carries a Credential name (for your own identification) and a Key prefix — a short identifier shown in the UI so you can recognize the token without revealing it. Use the show/hide toggle next to the bearer token field when you need to copy it.

Multiple Connections

You can create multiple MCP connections per user (e.g., one for your laptop, one for a CI pipeline). Each gets its own bearer token.

Scope & Security

	Detail
Organization scope	Every tool call is scoped to the organization the token belongs to. Claude can never see another workspace's data.
Bearer token auth	HTTP `Authorization: Bearer pb_...` header on every request
Token storage	Tokens are SHA-256 hashed and encrypted at rest. We never log raw tokens.
Read + write	MCP tools can create and update content. Regenerate the token immediately if exposed.
Audit trail	Errors and key activity are logged on the server side for support investigation.

Treat your MCP bearer token like a production API key — never paste it into chats, screenshots, or commits. Rotate it immediately if it ends up somewhere it shouldn't.

Troubleshooting

Claude Desktop: confirm you restarted the app after editing claude_desktop_config.json
Claude.ai: check the Connected apps page — the connector should show as active. Re-add it if not.
Config syntax: validate the JSON snippet — a missing comma or quote silently breaks tool discovery
Token: make sure the bearer token starts with pb_ and matches exactly what ProductBridge generated (no leading whitespace)

The bearer token isn't valid. Either:

You pasted a different token (check the Authorization header in your config)
The token has been regenerated — generate a fresh one and update your config
The token was revoked manually from the connections list

Be explicit in your prompt — "Use the ProductBridge MCP server to ..." helps Claude pick the right tools when it's juggling several. After a couple of turns, Claude usually starts reaching for them without prompting.

Check the error message — it usually tells you what's missing:

Status / category / tag not found — call the matching list_* tool first to get valid IDs
Board not specified — provide a board_id or rely on the workspace default
Invalid date format — eta on roadmap items must be YYYY-MM-DD

Was this page helpful?

Last updated 4 days ago

Built with Documentation.AI