toolchestrator

The Python client for Toolchestrator, the runtime for the small apps your team builds with Claude Code. It connects an app to its Toolchestrator runtime so teammates and AI can use it, while the app keeps running on the machine it was built on.

With this client a connected tool can:

• register itself in the company registry (name, author, the problem it solves);
• publish versioned JSON-Schema data models and sync records to the hub;
• serve remote task actions over an outbound long-poll (the hub never connects into your machine);
• read other tools' data through the hub with filtered queries;
• invoke another tool's action and get the result back (tc.call);
• expose its own local web UI through the hub (tc.serve_web).

Works on Python 3.9+; depends only on requests and websocket-client.

Install

pip install toolchestrator

Quick start

from toolchestrator import Toolchestrator

# one-time registration, with a personal token from your hub's Settings page
tc = Toolchestrator.register(
    "https://hub.example.com",        # your company's hub URL
    "tcu_...",                        # personal access token
    "invoice-radar", "Invoice Radar",
    description="Tracks vendor invoices and flags overdue ones.",
    sharing_scope="private",          # start private; widen deliberately
)

# afterwards, anywhere in the tool (loads .toolchestrator.json):
tc = Toolchestrator()
tc.sync_schema("invoice", invoice_schema)
tc.sync_data("invoice", rows, id_field="id")
tc.serve({"mark_paid": handle_mark_paid})   # execute tasks queued on the hub

The easiest way to connect a tool is to let an AI coding agent do the wiring: point it at your hub's connect guide (<your-hub-url>/skill) and ask it to "connect this tool to Toolchestrator". The hub URL always comes from you - the client never assumes one.

License

MIT © Stevica Kuharski