Set up coding agents

Choose capabilities

Vex keeps three agent integrations distinct:

• Skills provide durable guidance for safe Vex workflows.
• MCP tools let the agent inspect and act on live repository state.
• Agent Context is opt-in project-hook support that captures context, applies privacy checks, and attaches it to a commit checkpoint.

Choose the capability set that matches the harness and repository. An MCP tool install is not implied by a skill install.

Install and inspect

Run the setup command from the repository you want Vex to manage. Start interactively, or name targets and capabilities for a repeatable setup.

vex setup
vex setup --yes --targets codex,claude-code,cursor --with-mcp
vex setup --list-targets --format json
vex setup doctor --format json

Use vex setup refresh and vex setup uninstall --targets <target> to manage Vex-owned bundles. Modified bundles are preserved unless a command explicitly asks to force removal.

Agent Context

Agent Context keeps capture local and intentional. Use the doctor commands to inspect the host and remote readiness, then use checkpoint and resume paths when a repository workflow calls for them.

vex agent-checkpoints doctor
vex session doctor --remote --format json
vex session privacy-check <local-text-file> --format json

Do not treat context capture as a public upload of every transcript. The workflow applies configured privacy checks and targets the exact committed work.

Machine-readable docs

Human documentation lives here. Vex also serves separate versioned Markdown and JSON endpoints for agents:

• /skill.md
• /skills/index.json
• /docs/import/prompt.md

Those endpoints have stable machine-oriented contracts and remain separate from the public docs navigation.