Claude Code Integration Guide

Open Source Projects

Community-built projects that wrap Claude Code CLI for programmatic use:

Project	Description	Tag
using-claude-code-cli-agent-skill	Claude Code skill for programmatically invoking Claude Code CLI from Python orchestrators, shell scripts, and automation pipelines. https://github.com/SpillwaveSolutions/using-claude-code-cli-agent-skill	Best for servers
openclaw-claude-code	Turns Claude Code CLI into a programmable headless engine with subprocess management, agent teams, and cost tracking. https://github.com/Enderfga/openclaw-claude-code	Multi-agent
ralph-claude-code	Autonomous loop that continuously runs Claude Code with automatic session management across iterations. https://github.com/frankbria/ralph-claude-code	Autonomous loop
learn-claude-code	Minimal Python agent harness built from scratch. Great for understanding how Q&A loops work under the hood. https://github.com/shareAI-lab/learn-claude-code	Educational
CheetahClaws	Python-native Claude Code reimplementation — lightweight, readable, supports multiple models. https://github.com/SafeRL-Lab/nano-claude-code	Python-native
claude-agent-sdk-python	Official Anthropic Agent SDK for Python. Runs in-process — no subprocess management needed. https://github.com/anthropics/claude-agent-sdk-python	Official SDK

Access Options Overview

Method	Use Case	Requires
Claude Agent SDK	Build apps with agentic Claude	Anthropic API key
Anthropic API	Raw model access, full control	Anthropic API key
CLI Subprocess	Script / automate Claude Code	Claude Code + Pro or API key
MCP Server	Give Claude tools from your app	Claude Code installed

Claude Pro subscription ($20/month on claude.ai) does not grant API access. To use the Agent SDK you need a separate account at console.anthropic.com. However, the CLI subprocess approach can use your Pro subscription via claude login.

CLI Subprocess — Ask / Answer Loop

Claude Code supports non-interactive scripting mode. Use -p to run a single query and exit, then carry the session_id forward to maintain conversation context.

Key Flags

Flag	Description
`-p / --print`	Non-interactive mode — run and exit
`--output-format json`	Structured JSON output
`--output-format stream-json`	Streaming newline-delimited JSON
`--resume <session_id>`	Resume a previous session
`--allowedTools`	Pre-approve tools, skip prompts
`--cwd <path>`	Set working directory / project folder

Getting a session_id

The session_id is returned automatically in the first JSON response — you do not create it manually.

claude -p "Hello" --output-format json

# Response:
{
  "result": "Hello! How can I help you?",
  "session_id": "55ac75f5-3c3e-4e24-9e39-bf6401e4456c",
  ...
}

Note: uuid in the response is the message/request ID — not the session. Always use session_id for --resume.

Python Q&A Loop

import subprocess, json

session_id = None
print("Claude Q&A Loop (type 'exit' to quit)\n")

while True:
    question = input("You: ").strip()
    if question.lower() == "exit":
        break

    cmd = ["claude", "-p", question, "--output-format", "json"]
    if session_id:
        cmd += ["--resume", session_id]

    result = subprocess.run(cmd, capture_output=True, text=True)
    data = json.loads(result.stdout)

    session_id = data.get("session_id")
    print(f"Claude: {data['result']}\n")

Point Claude at a project folder

# Option 1 — cd into the folder
cd /path/to/project
claude -p "Summarize this project" --output-format json

# Option 2 — use --cwd flag
claude -p "Summarize this project" --cwd /path/to/project --output-format json

# Option 3 — from Python subprocess
subprocess.run(
    ["claude", "-p", "Summarize this project", "--output-format", "json"],
    cwd="/path/to/project",
    capture_output=True, text=True
)

Claude Agent SDK

Gives you Claude Code's agentic capabilities as a library — no subprocess management needed.

Install

# Python
pip install claude-agent-sdk

# TypeScript / Node.js
npm install @anthropic-ai/claude-agent-sdk

Concurrent Q&A Server (TypeScript)

import Anthropic from "@anthropic-ai/sdk";
import express from "express";

const client = new Anthropic();
const app = express();
app.use(express.json());

app.post("/ask", async (req, res) => {
  const session = await client.beta.sessions.create({});

  const response = await client.beta.messages.create({
    model: "claude-sonnet-4-6",
    max_tokens: 1024,
    session_id: session.id,
    messages: [{ role: "user", content: req.body.question }],
  });

  await client.beta.sessions.delete(session.id);
  res.json({ answer: response.content[0].text });
});

app.listen(3000);

CLI vs Agent SDK

	CLI Subprocess	Agent SDK
Claude Pro support	Yes (via claude login)	No (API credits only)
Speed	Slower (new process each turn)	Fast (in-process)
Multi-turn loop	Yes, via --resume	Yes, natively
Best for	Scripting, low frequency	Production servers

MCP — Model Context Protocol

MCP lets your program expose tools that Claude Code can call. Your program runs as an MCP server; Claude connects and calls your functions automatically.

Use MCP when you want Claude to call APIs, databases, or functions in your app.
Use the Agent SDK when you want your app to drive Claude as an agent.

Configure MCP servers in .claude/settings.json.

Which Approach Should I Use?

Your program wants to... │ ├── Give Claude tools (files, APIs, DBs) │ └── Build an MCP Server │ ├── Control Claude as an agent from code │ └── Use the Agent SDK │ ├── Simple scripting / CI / use Pro subscription │ └── Use CLI Subprocess (-p flag) │ └── Full low-level control over every token └── Use the Anthropic API directly

Authentication

CLI — use your Pro subscription

claude login

Agent SDK / Anthropic API — use an API key

export ANTHROPIC_API_KEY="sk-ant-..."

Get an API key at console.anthropic.com