What is toli?
toli connects AI agents so they can delegate tasks, ask questions and return results to one another, even when they use different agent harnesses or run on separate machines. It is aimed at people coordinating coding agents, personal assistants or custom agents, with each agent connected through MCP and agent-to-agent communication handled through A2A. You run a daemon and choose which agents may work together, then manage their tasks from a terminal interface or CLI.
toli Features:
Local Daemon: A daemon runs on macOS, Linux or Windows and gives each registered agent a name and token for connecting through MCP.
Harness Connections: Built-in connection commands support Claude Code, Codex, OpenCode, Pi and OpenClaw, while other agents can connect through an MCP client.
A2A Handoffs: Agents and their approved peers use the A2A protocol to delegate tasks, ask questions and send results back.
Peer Pairing: An operator adds peers using the CLI or TUI, with one-time invitations or tokens used to establish connections between agents.
Task Monitoring: The terminal interface and CLI let you inspect delegated tasks and view conversations between agents.
Agent Notifications: When a task arrives, toli can wake an agent through a harness connection, a pull listener, a webhook or a program.
toli Benefits:
Shared Work: Agents built for different harnesses can contribute to the same workflow instead of requiring you to keep each one isolated.
Less Context Switching: An agent can ask a peer for information or hand off a task, reducing the need to relay every detail between separate sessions yourself.
Clear Oversight: A central view of task conversations makes it easier to follow what agents have requested and returned.
Controlled Access: Because you choose the peers and agents cannot add peers themselves, you retain control over which agents may collaborate.
Flexible Setup: Local and networked deployments let teams coordinate agents on one computer or across machines, using a network setup they manage.
Use Cases:
Frontend and API Work: Pair a web-interface agent with an API agent so one can ask the other about its current work or delegate a related task.
Remote Agent Teams: Connect agents running on a teammate’s machine to collaborate across a network, with each operator pairing the peers they trust.
Assistant Research: Let a personal assistant check with a coding agent before answering a question that depends on the state of a software project.
Task Supervision: Use the terminal interface or CLI to review the exchanges when several agents are delegating work and returning results.
Custom Agent Workflows: Connect an MCP-enabled agent or service so it can receive delegated tasks and respond through a pull listener or webhook.

