Opencode MCP Client Explained
Looking for a complete overview? → Read the MCP Hub: /en/mcp
The Opencode MCP client is the runtime layer responsible for executing MCP logic.
This page explains how the client works and when customization is necessary.
What Is the MCP Client?
The MCP client:
- Sends structured context to the model
- Receives model responses
- Executes tool calls
- Updates MCP state
It acts as the bridge between protocol definition and real execution.
MCP Client Responsibilities
The client is responsible for:
- Context serialization
- Tool dispatch and isolation
- Error handling
- State reconciliation
This separation keeps agent behavior deterministic.
Client–Server Interaction
Typical interaction flow:
- Client sends context snapshot
- Server validates and processes context
- Tools execute in isolated environments
- Results return to the client
- Context is updated and persisted
Do You Need a Custom MCP Client?
Most users do not.
You only need a custom client if:
- Embedding Opencode into another system
- Implementing custom tool routing
- Requiring non-standard persistence behavior
Performance Considerations
- Smaller context sizes improve performance
- Limit tool calls per step
- Avoid unnecessary global state
Related Guides
Summary
The MCP client turns protocol design into working agents.
Understanding it helps you debug, optimize, and extend Opencode safely.