Before Mission Control can dispatch tasks, you need to configure at least one agent. An agent configuration tells Mission Control how to reach a specific AI tool, what credentials it needs, and how it should behave.
Mission Control currently supports three agent providers:
Claude Code — Anthropic’s AI coding agent. Requires an Anthropic API key. Excellent at reasoning through complex, open-ended tasks with minimal guidance.
Codex CLI — OpenAI’s code-focused CLI agent. Requires an OpenAI API key. Works well for well-scoped, well-specified changes where speed is a priority.
Gemini CLI — Google’s coding agent with an exceptionally large context window. Requires a Google AI API key. Useful when tasks require analyzing large files or entire modules at once.
Navigate to Settings > Agents and click Add Agent. The configuration form includes:
Name — A human-readable label for this agent profile. Use something descriptive like “Claude Code - Primary” or “Codex - Fast Tasks”.
Provider — Select from the supported provider list.
API Key — The credential used to authenticate with the AI provider. API keys are stored encrypted in the local SQLite database and never sent to any external service.
Working Directory — The filesystem path where this agent should operate. Typically the root of your project repository.
Timeout — How long Mission Control waits for an agent response before marking the task as failed. Default is 300 seconds; increase this for tasks involving large codebases.
After saving an agent profile, use the Test Connection button to verify Mission Control can reach the agent successfully. This sends a lightweight ping through the gateway and confirms the API key is valid.
If the test fails, check that:
You can create multiple profiles for the same provider with different configurations. For example, you might have one Claude Code profile pointed at your main project and another pointed at a separate documentation repository.