drydock-sdk

Python SDK for drydock — build and test AI agents and agentic workflows locally, one binary, with or without Docker.

Write an agent, tool, or MCP server as a plain Python function, decorate it, and the drydock runtime launches it as an isolated node: hot-reloaded during development, wired to other agents (A2A) and MCP servers, and fully traced.

from drydock import node

@node(name="greeter", role="agent")
async def greet(ctx, name: str = "world"):
    await ctx.emit({"status": "thinking"})   # stream a partial update
    return {"message": f"hello, {name}"}

Requires the drydock runtime

This package is the author-side SDK only. It has no effect on its own — nodes are launched and orchestrated by the separate drydock binary (the runtime, HTTP API, web UI, and tracing all live there). Install the runtime first:

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/scionoftech/drydock/main/scripts/install.sh | sh

See the project README for tarball and build-from-source options.

Install

pip install drydock-sdk

Zero runtime dependencies, pure Python, ships type hints (PEP 561). Install it into the virtualenv of each node you author.

Write a node

A node is a directory with your handler and a node.yaml manifest:

# handler.py
from drydock import node

@node(name="hello", role="agent")   # role: "agent" | "tool" | "mcp"
async def hi(ctx, name: str = "world"):
    return {"message": f"hello, {name}"}

# node.yaml
name: hello
version: 0.1.0
role: agent
runtime: python
entry: handler.py:hi

drydock dev                                   # hot-reload daemon + web UI
drydock invoke hello --input '{"name":"Sai"}'

The ctx object

Every handler receives a context as its first argument:

Handlers may be async def (served concurrently) or plain def (run in a thread pool). A JSON-object input is spread onto handler kwargs; return any JSON-serializable value.

Full reference — input mapping, session semantics, the error model, streaming, and environment scrubbing — is in docs/SDK.md.

Links

• Project & docs: https://github.com/scionoftech/drydock
• SDK reference: https://github.com/scionoftech/drydock/blob/main/docs/SDK.md
• Issues: https://github.com/scionoftech/drydock/issues

License

MIT