Claude Code Setup¶

Set up Claude Code to generate idiomatic adk-fluent and Google ADK code.

1. Project rules — `CLAUDE.md`¶

Claude Code reads CLAUDE.md from your project root automatically at the start of every session.

If you cloned the repo¶

The file is already generated at CLAUDE.md in the project root. Nothing to do — Claude Code picks it up automatically.

CLAUDE.md is auto-generated by scripts/llms_generator.py from the ADK manifest. It updates automatically when the API changes via just docs or CI. Do not edit it manually.

If you installed via pip¶

Download the rules file into your project:

curl -L https://raw.githubusercontent.com/vamsiramakrishnan/adk-fluent/master/CLAUDE.md \
  -o CLAUDE.md

Quick use — llms.txt¶

You can also reference the hosted llms.txt directly in a conversation:

#fetch https://vamsiramakrishnan.github.io/adk-fluent/llms.txt

2. MCP server — live documentation access¶

MCP servers give Claude Code on-demand access to the full adk-fluent documentation, including API references, cookbook recipes, and method signatures.

Option A: adk-fluent GitMCP (free)¶

Uses GitMCP to serve documentation directly from the GitHub repository.

claude mcp add --transport http adk-fluent https://gitmcp.io/vamsiramakrishnan/adk-fluent

No authentication required. Claude Code will automatically have access to all adk-fluent documentation.

Usage — just ask naturally:

Build me a pipeline with a researcher agent that searches the web,
then a writer agent that summarizes the findings using adk-fluent.

Option B: Context7 MCP¶

Community MCP server that provides documentation context for many libraries including adk-fluent.

claude mcp add --transport http context7 https://mcp.context7.com/mcp

claude mcp add context7 -- npx -y @upstash/context7-mcp

Usage — append use context7 to your prompt:

Build me a pipeline with a researcher and writer agent using adk-fluent. use context7

3. Verify the setup¶

After adding CLAUDE.md and an MCP server, test it with a prompt like:

Create an adk-fluent agent that classifies customer support tickets
into categories, then routes them to specialized handler agents.
Use a FanOut for parallel processing and write results to state.

Claude Code should:

Import from adk_fluent (not internal modules)
Use the fluent builder pattern with method chaining
Call .build() to produce native ADK objects
Use .writes() for state management
Follow the project conventions from CLAUDE.md