Extensions

Eyeban does not ship a public extension marketplace. Extend the agent with MCP servers, hooks, rules, and skills you control.

MCP configuration

MCP supports stdio and SSE / streamable HTTP. Configure servers in:

• User scope: ~/.eyeban/mcp.json
• Project scope: .eyeban/mcp.json

Project scope overrides user servers with the same name. Dedicated how-to: the MCP docs page.

Project scope is useful when the whole team should share the same tool wiring. Do not commit tokens or secrets into these files.

Hooks

Lifecycle hooks live in hooks.json (project or user location depending on your setup). Use hooks for policy checks, logging, or gating risky tool classes — not as a substitute for permissions Ask / Allow safe / Allow all.

Rules and skills

• Rules: .eyeban/rules/*.mdc — durable instructions injected into agent context
• Skills: .eyeban/skills — packaged procedures the agent can follow for repeatable workflows

Optional legacy instruction files from other editors may be present in a repo; prefer .eyeban/ paths as the supported surface.

Trust boundary

• Prefer MCP servers you can audit
• Pin versions when possible
• Review tool schemas before Allow all
• Do not install third-party “Eyeban extensions” from untrusted sites

What is not supported yet

There is no curated store, signed extension packages, or automatic sync of third-party plugins. Treat configuration-as-code in your repository as the extension model.