Agent Workflows

The use case

AI agents (Claude Desktop, Cursor) need access to changelog data to answer questions like "What did Vercel ship last week?" or "Show me breaking changes in the Stripe API."

How the MCP server works

Watch Changelog's MCP integration is a local stdio (Model Context Protocol) process, not a hosted remote service. Signed-in customers can create API keys in /app/api-keys and use them with local MCP or the REST API.

The MCP server provides six tools:

• list_sources lists sources selected for the API key, with optional vendor, search, and pagination filters.
• get_entries returns normalized entries for selected source slugs, with optional date and pagination filters.
• get_my_updates, get_interests, and save_interests read and manage saved source interests.
• request_source submits a website or GitHub repository for review.

Setup

Free covers local MCP and REST for 10 followed sources. Pro includes unlimited catalog access.

Install Node.js 20 or newer, then download the standalone client to a path you control. It runs locally over stdio and defaults to the Watch Changelog API.

mkdir -p "$HOME/.local/share/watchchangelog"
curl -fSL https://watchchangelog.com/mcp/watchchangelog.mjs \
  -o "$HOME/.local/share/watchchangelog/watchchangelog.mjs"
node "$HOME/.local/share/watchchangelog/watchchangelog.mjs"

In Claude Desktop or Cursor, use this configuration and replace the placeholder locally:

{
  "mcpServers": {
    "watch-changelog": {
      "command": "node",
      "args": ["/absolute/path/to/watchchangelog.mjs"],
      "env": {
        "WATCHCHANGELOG_API_KEY": "wcl_your_api_key"
      }
    }
  }
}

Example prompts

• "List my selected sources."
• "Show entries from my selected Stripe source since 2026-08-01."
• "Save React and Next.js as my source interests."
• "Show my updates from those saved interests."

The agent uses the local MCP server to fetch data and format the answer in natural language.