<aside> 🧭

Audience: Backend engineers building or adapting MCP servers

This guide provides resources to help you start building an MCP server to expose MCP Tools and make them accessible to our agents

</aside>

Connection and Authentication Requirements

• Transport: Streamable HTTP only (no SSE or stdio)
• Authentication: None, Token (header-based), or OAuth.
• Capabilities: Tools only (resources and prompts not supported)
• Hosting: Your MCP server must be hosted on the infrastructure you control and be reachable on the public internet over HTTPS.

Building a MCP server step by step

For a step-by-step tutorial on how to build an MCP server, please refer to the official documentation: Build an MCP server — Model Context Protocol

Testing Your Server

A simple way to test your server locally is by using the MCP Inspector (GitHub)

# install node lts and execute 
npx @modelcontextprotocol/inspector

Integration

• Once you connect your MCP, you cannot change its URL. Any new URL must be added as a new MCP.

• Updating your MCP:

  • Deleting a tool: Remove all tool references in the Typewise platform before deleting an MCP tool from your server. Failing to do this can cause AI Agents that reference the tool to handoff, even when the tool isn't invoked.
  • Modifying existing tools' signatures or return types may confuse agents using them and trigger a handoff. Avoid such modifications whenever possible. If changes are necessary, be sure to update the "Action description" by reverting it to the new default.

  

  • Adding new tools is fully supported. You can immediately add them to Knowledge or Action Hub once they become available.