7. Tools, Tool Calling and MCP

7.1 Why tools matter

An LLM alone can generate text.

An agent with tools can interact with an environment.

Model
 |
 +--> Read files
 +--> Search files
 +--> Edit files
 +--> Run commands
 +--> Run tests
 +--> Search web
 +--> Use external integrations

Conceptually:

Model
 |
 | "I need to run tests"
 v
Tool request
 |
 v
Runtime executes tool
 |
 v
Tool result
 |
 v
Model continues

The model does not magically execute a command. The surrounding application/runtime controls tool execution.

MCP stands for Model Context Protocol.

It is a protocol used to connect AI applications with external tools and data sources.

Conceptually:

OpenCode
   |
   v
MCP server
   |
   +--> GitHub
   +--> Database
   +--> Internal API
   +--> Documentation

The exact MCP integration depends on the server and OpenCode configuration.

Tools can be powerful.

A tool that can:

delete files
push code
modify infrastructure

should not necessarily be unrestricted.

Use least privilege and approval gates.

For each capability below, classify it as:

Read source
Run tests
Git status
Git diff
Git commit
Git push
Delete production data

Explain your choices.