Paragon MCP Server – a Node/TypeScript implementation of the Model Context Protocol that exposes Paragon ActionKit integrations (130+ SaaS actions) as MCP tools, with optional support for custom OpenAPI actions and an experimental proxy API tool.
https://github.com/useparagon/paragon-mcpBuilding AI agents that actually connect to your users' SaaS tools? You know the drill: spend weeks wiring up OAuth flows, parsing API responses, and handling rate limits for each service. Then multiply that by however many integrations your users need.
Paragon MCP cuts through this entirely. It's an MCP server that gives your AI agents direct access to 130+ pre-built SaaS actions through Paragon's ActionKit API. No more building Salesforce connectors from scratch or wrestling with Google Calendar's quirky date handling.
Your AI agent needs to create calendar events? Update CRM records? Send Slack notifications? Those are table stakes now, not competitive advantages.
Instead of spending engineering cycles on integration plumbing, you get:
Each integration comes with proper OAuth handling, error management, and API best practices baked in. The kind of stuff that takes weeks to get right when you're building it yourself.
Here's where it gets interesting. The MCP includes Paragon's Connect Portal - a pre-built OAuth component that handles the authorization dance for your users. They connect their accounts once through a secure flow, and your agent gets persistent access to their data.
No more:
The Connect Portal handles all of this. Your users see a clean interface, authorize their accounts, and you get authenticated API access immediately.
Customer Support Agent: When a user reports an issue, your agent automatically creates a ticket in Zendesk, adds the customer to a Slack channel, and schedules a follow-up in Google Calendar. Three different SaaS tools, one conversation.
Sales Workflow Automation: Lead comes in through your website, agent creates a Salesforce record, sends a personalized Slack message to the sales rep, and adds a follow-up task in Asana. All triggered by natural language requests.
Content Management: "Update our project timeline and notify the team." Your agent updates the Notion database, posts to the relevant Slack channel, and creates calendar reminders for upcoming deadlines.
The MCP doesn't lock you into just the included actions. Drop OpenAPI specs into the openapi/ folder, and those endpoints become available tools automatically. Need custom Spotify integration? Define the spec, and your agent can use it.
There's also an experimental proxy API tool that lets your agent write raw HTTP requests to any connected service. Risky but powerful for edge cases where pre-built actions don't cover your needs.
Adding this to your existing setup is straightforward:
For Cursor:
{
"mcpServers": {
"paragon-actionkit": {
"url": "http://localhost:3001/sse?user=[user-id]"
}
}
}
For Claude:
{
"mcpServers": {
"actionkit": {
"command": "npx",
"args": ["mcp-remote", "http://localhost:3001/sse?user=[user-id]"]
}
}
}
The server runs locally for development, but you'll want to deploy it for production use with proper user authentication via JWT tokens.
You're not building a calendar app or a CRM. You're building something that needs to work with calendars and CRMs. Paragon MCP lets you focus on your core logic while having reliable, tested integrations available as simple function calls.
The time you save not building individual integrations? That's time you can spend on the features that actually differentiate your product.
Your users get native integration experiences. Your agents get powerful capabilities. You get to ship faster.