How do I connect a custom MCP to Pylon?

Last updated: August 14, 2026

You add your own internal tools as MCP connections into Pylon for use within Pylon's AI features like Ask AI or the Issue Copilot.

Setup

  1. Go to the MCP section of the Integrations page.

  2. Click on Add MCP Connector

  3. Fill in the required details based on either the third party's MCP documentation or your own internal MCP details.

The Icon, Name, and Description will be visible when members of your org are using the App.

  1. Each user of your organization will now see the app on their integrations page. For OAuth-based MCP connections, users will need to individually connect their account by clicking the app and hitting Connect.

Authentication

Pylon's MCP client connects over Streamable HTTP and implements the MCP authorization spec (2025-06-18), using PKCE (RFC 7636) on every flow and discovering OAuth endpoints via Protected Resource Metadata (RFC 9728) and Authorization Server Metadata (RFC 8414).

A connector can obtain its OAuth client in one of two ways:

  • Dynamic Client Registration — if the MCP server advertises a registration endpoint per RFC 7591, Pylon registers itself automatically on the first connect, so no Client ID is required.

  • Manual Client ID — for servers that don't support dynamic registration, register an OAuth app with the provider yourself and paste the Client ID and optionally Client Secret into the connector. If the connection doesn't work with the Client Secret value, try removing it in both places: the OAuth app registration page, and the Pylon configuration settings.

In either case, each user connects their own account, and tokens are stored encrypted and refreshed automatically.

Can I connect a public MCP server with no authentication?

No. Custom MCP connectors in Pylon currently expect an OAuth-capable MCP server. The server must expose the OAuth metadata Pylon needs and either support dynamic client registration or have a client ID configured in Pylon.

If the MCP server does not require authentication and does not expose those OAuth endpoints, the Connect flow cannot generate an authorization URL and the connection will fail.

What to do instead

  • If your goal is to make public documentation searchable in Pylon AI features, add the public site as Training Data instead.

  • If the system does not expose an MCP server you can use, consider a Custom API tool for supported API-based workflows.

Usage

Users can mention Apps in natural language and queries will automatically attempt to pull from your connect MCPs where relevant.

Users can also see which MCPs are connected and in-use from the Apps menu.