Calculator

You can connect your local language model agent to a centralized cloud MCP server that exposes calculator capabilities via Server-Sent Events. This setup enables students to perform real-time computations through a shared tool managed by an instructor, while your local agent handles requests and displays results seamlessly.

How to use

Set up and connect using an MCP client in your local environment. Your client will open a persistent HTTP-SSE connection to the cloud calculator server, initialize the session, and then issue tool calls to perform calculations. The cloud server processes expressions and returns results that your agent can present to you.

How to install

Prerequisites vary by your local setup, but you typically need Docker on the cloud instructor VM and a client capable of MCP with SSE support on your machine.

On the instructor side, use the provided command flow to build and run the cloud service, then ensure port 8080 is reachable from student machines.

# On the instructor VM
make build
make run
# Check that the server is listening on port 8080
curl -v http://YOUR-VM-PUBLIC-IP:8080/sse

Additional notes and setup details

Student side configuration involves updating your local MCP client configuration to point at the cloud server, using a dedicated MCP server entry. In this example, a cloud-calc server entry is defined with an HTTP endpoint.

Use the following sample to configure your MCP client. Replace YOUR-VM-PUBLIC-IP with the actual public IP address of the instructor's VM.

mcpServers:
  - name: cloud-calc
    type: http
    url: http://<YOUR-VM-PUBLIC-IP>:8080/sse
    args: []

experimental:
  autoExecuteTools: true

Troubleshooting and security notes

If connections fail, verify the cloud firewall on the instructor VM allows inbound TCP traffic on port 8080. Review server logs on the VM to identify connection or startup issues.

For debugging from the cloud server side, you can follow the live server logs with the Makefile target.

Available tools

calculator

Evaluates mathematical expressions and returns computed results as JSON-RPC responses through the MCP SSE channel.