Skip to main content
AgentCall lets one person call another person’s Claude Code or Codex agent from the terminal. The owner chooses what their agent can do; the caller sends a message to an address such as @acme/ken and receives the answer on standard output.
AgentCall is friends-scale software, not a hostile-code sandbox. The default task delegates the owner’s connected MCP servers, skills, and web tools. Review that authority before making an agent callable.

Call an agent

Install AgentCall, enroll, and make a first authenticated call.

Answer calls

Keep a listener online and review which authenticated tools callers inherit.

Understand the boundary

See what is encrypted, what the relay observes, and what remains risky.

​
What happens during a call

  1. The caller selects an address, optional task, and message.
  2. AgentCall verifies the callee’s signed key record and encrypts the request.
  3. The relay authenticates both sides and routes ciphertext inside one organization.
  4. The callee resolves task policy before the caller’s message enters the prompt.
  5. A fresh Claude Code or Codex process answers in the configured working directory.
  6. The listener encrypts the outcome and the caller prints the authenticated reply.
Read how AgentCall works for the lifecycle and core concepts for the vocabulary.

​
Choose your path

Source of truth: repository overview.