MCP Server
Query ProductSights from Claude Code, Cursor, VS Code, and Windsurf.
ProductSights provides a Model Context Protocol (MCP) server that lets you query your feedback data directly from AI coding tools — without leaving your editor.
Supported IDEs
- Claude Code (CLI)
- Cursor
- VS Code (with Copilot or compatible extension)
- Windsurf
Installation
Install and run the MCP server using npx:
npx @productsights/mcp-server
The server requires an API key. Generate one from Settings → API Keys in your ProductSights dashboard.
Available tools
The MCP server exposes tools your AI assistant can call:
| Tool | Description |
|---|---|
| search_insights | Search insights by keyword, category, or sentiment |
| get_top_problems | Get top user problems grouped into clusters |
| get_insight_stats | Get analytics summary across all feedback |
| get_weekly_summary | Get a weekly summary of feedback activity and trends |
| get_recent_insights | Fetch latest insights with optional filters |
| find_related_insights | Find feedback related to any topic |
| find_insights_by_meaning | Semantic search — find insights by meaning using vector embeddings, not just keywords |
| submit_insight | Create a new insight directly from your IDE |
| correct_insight | Fix an AI triage classification (category, sentiment, priority) for a specific insight |
| get_triage_accuracy | Monitor how often AI triage is corrected — useful for tracking classification quality |
| get_weekly_decision_packets | Retrieve weekly PM/engineering decision packets for top clusters |
| get_cluster_impact | Retrieve before/after impact metrics for a specific cluster |
| get_cluster_detail | Get full context for a cluster — evidence, discovery state, and signal metrics |
| update_cluster_status | Set execution status or assign an owner to a cluster |
| get_spec_context | Get product context optimized for writing specs — top problems, trends, and cluster summaries |
Example usage
Once connected, you can ask your AI coding assistant questions like:
- "What are the top bugs users are reporting?"
- "Search for feedback about the login flow"
- "What's the sentiment around our API?"
- "Find insights related to the feature I'm working on"
- "Submit an insight about the slow dashboard load time"
- "What's the triage accuracy this week?"
- "Get context for writing a spec on the onboarding flow"
The AI assistant will call the appropriate MCP tool and present the results inline in your editor.
Bidirectional MCP
ProductSights also supports using any MCP-compatible server as a feedback source. This means you can pipe output from external MCP tools into ProductSights:
- Go to Settings → Sources and add an MCP Server source
- Configure the server command, arguments, and tool to call
- Choose an output mapping mode:
- Raw — treat the entire output as one feedback item
- JSON Array — parse the output as an array of items
- JSON Path — extract items using a JSON path expression
- Map fields (content, id, timestamp) from the output to ProductSights fields
This is useful for collecting feedback from custom internal tools, databases, or third-party APIs that expose an MCP interface.