# Claude Agent SDK vs Google ADK

> Source: https://theagentsindex.com/compare/claude-agent-sdk-vs-google-adk (structured, researched, re-verified)

The Claude Agent SDK and Google Agent Development Kit (ADK) are two of this category's four official, major-lab agent SDKs, alongside OpenAI Agents SDK and Microsoft Agent Framework. Both listings on this site already treat each other as a close comparison: google-adk's alternatives rail names claude-agent-sdk fourth, and claude-agent-sdk's rail reciprocally names google-adk second.

They arrived from opposite design goals. The Claude Agent SDK is not a general-purpose framework that happens to support Claude. It is the literal agent harness behind Claude Code (the same tool loop, context management and toolset), exposed as a Python or TypeScript library via a one-shot `query()` function or a stateful `ClaudeSDKClient`.

Google ADK is a from-scratch product, announced at Google Cloud NEXT in April 2025, built protocol-first around interoperability and evaluation tooling instead of wrapping an existing CLI.

Openness and language spread separate them sharply. The Claude Agent SDK's Python and TypeScript wrapper packages carry an MIT license, but Anthropic's own docs state that use of the SDK as a whole is governed by its Commercial Terms of Service. The compiled Claude Code CLI binary the wrapper bundles and spawns as a subprocess isn't published as source: it doesn't clear this index's bar for a fully self-hostable open-source tool, and it ships in two languages only.

Google ADK is Apache 2.0 at the core, genuinely self-hostable anywhere for free, and ships production-grade SDKs in four languages (Python, Go, Java, TypeScript) plus an experimental Kotlin/Android runtime, the broadest spread of any framework in this index.

Checked live via the GitHub API on 2026-08-07: google/adk-python carries 21,029 stars and 3,809 forks, versus a combined 9,504 stars across the Claude Agent SDK's two language repos (anthropics/claude-agent-sdk-python: 7,826 stars / 1,212 forks; anthropics/claude-agent-sdk-typescript: 1,678 stars / 208 forks). All three repos were pushed within the last day of this research, real active projects.

The model-provider story follows the same split. The Claude Agent SDK runs Claude models only; a GitHub issue asking Anthropic to support other providers was closed "not planned". ADK is optimized for Gemini but genuinely reaches OpenAI, Anthropic Claude, Cohere, local Ollama models and 100+ others through a LiteLLM integration.

Both support the Model Context Protocol, but only ADK ships native Agent2Agent (A2A) protocol support for cross-vendor agent interoperability. A2A is now Linux Foundation-governed, with 150+ organizations reported in production; the Claude Agent SDK's integrations list has no A2A equivalent.

Production tooling diverges too. The Claude Agent SDK ships hooks that intercept and approve or block specific tool calls, subagents for delegating focused subtasks, and in-process custom tools via an `@tool` decorator mounted as an SDK MCP server, avoiding external-server overhead for simple cases.

ADK ships a graph-based Workflow Runtime (sequential/parallel/loop/conditional routing with human-in-the-loop steps), a Task API for structured agent-to-agent delegation, a local ADK Web debugging UI, and `adk eval`, a repeatable framework for scoring an agent's full trajectory and final output. Both offer a self-host-or-managed choice: the Claude Agent SDK's subprocess can run on your own Docker/Kubernetes/sandbox infrastructure or be handed to Claude Managed Agents, Anthropic's own hosted REST API; ADK's agent can self-host anywhere for free or deploy to the managed Agent Runtime on Google Cloud.

Pricing is free at the SDK layer on both sides, with the bill landing on model usage or managed hosting instead. The Claude Agent SDK is a free download; Claude API usage is billed per model. Haiku 4.5 runs $1/$5 per million input/output tokens, Sonnet 4.6 runs $3/$15, Sonnet 5 runs an introductory $2/$10 through August 31, 2026 (standard $3/$15 from September 1), and Opus 5/4.8/4.7 run $5/$25.

Prompt caching (10% of input price on a cache hit) and a 50% Batch API discount are both available, re-confirmed live against Anthropic's own pricing page on 2026-08-07 with no drift from the listing. Routing the same workload through Claude Managed Agents adds $0.08 per session-hour, billed only while a session is actively running.

ADK's managed Agent Runtime instead meters $0.0864 per vCPU-hour plus $0.0090 per GB-hour, with a free allowance of 50 vCPU-hours and 100 GB-hours per month, plus $0.25 per 1,000 events for session and memory storage. No flat monthly fee either way.

One pricing signal favors ADK's stability. Anthropic changed how programmatic Claude usage is billed twice within about six weeks in 2026, reversing a planned split onto a separate "Agent SDK credit" pool on the day it was due to take effect (June 15) after user backlash. Google's per-unit Agent Runtime metering has been independently re-verified unchanged across two checks nine days apart (2026-07-29 and 2026-08-07).

| Fact | Claude Agent SDK | Google Agent Development Kit |
| --- | --- | --- |
| What it is | Anthropic's own agent-building library: the same agent loop, tools and context management that power Claude Code, for Python and TypeScript. Claude-only, with no multi-provider escape hatch. | Google's open-source, code-first agent SDK with five language runtimes (Python, TypeScript, Go, Java, Kotlin), native A2A and MCP support, and an optional managed runtime on Google Cloud. |
| Pricing | Free | Free |
| API | Yes | Yes |
| Tags | API · Claude · Free tier · Multi-agent · Self-hosted | API · Claude · Free tier · GPT · Multi-agent · Open models · Open source · Self-hosted |
| Best for | Developers already building on Claude (or Claude Code) who want its exact agent loop, tools and context management as a callable library, and are comfortable being single-vendor on Anthropic in exchange for the tightest possible integration with Claude Code's own capabilities. | Teams that want Google's own official, actively-developed agent SDK, especially polyglot teams (Python/Go/Java/TypeScript/Kotlin) or anyone building multi-agent systems that need to interoperate across vendors via the A2A protocol. |
| Full facts | https://theagentsindex.com/claude-agent-sdk.md | https://theagentsindex.com/google-adk.md |

## Verdict: Claude Agent SDK or Google Agent Development Kit?

Pick the Claude Agent SDK if you're already building on Claude or Claude Code and want its exact production harness, the same tools, agent loop and context management, as a callable Python or TypeScript library. Single-vendor commitment to Anthropic is an acceptable tradeoff for the tightest possible integration with Claude Code's own capabilities (hooks, subagents, in-process MCP tools).

Pick Google ADK if genuine multi-provider flexibility in one agent loop matters (Gemini, OpenAI, Claude, Cohere or local models via LiteLLM), if you need the broadest language spread in this category (production Python, Go, Java, TypeScript, plus an experimental Kotlin/Android runtime), or if native Agent2Agent protocol support for cross-vendor agent interoperability is a real requirement rather than a nice-to-have.

ADK is also the more conventionally open-source choice: Apache 2.0 end to end, versus the Claude Agent SDK's MIT-licensed wrapper around a closed-source, Commercial-Terms-governed CLI binary. It carries by far the larger, more active community too, with 21,029 GitHub stars versus a combined 9,504 across the Claude Agent SDK's two language repos.

Skip the Claude Agent SDK if Anthropic vendor lock-in is a real constraint, or if you need billing terms that have already settled; Anthropic changed Claude Agent SDK billing twice in six weeks in 2026. Skip ADK if your team is fully committed to Claude and has no need for multi-language or multi-provider reach. That tighter, single-vendor integration is the Claude Agent SDK's entire reason to exist.

## FAQ

**Are the Claude Agent SDK and Google ADK direct competitors?**

Yes — both listings on this site already name each other as a close comparison: Google ADK's alternatives rail describes the Claude Agent SDK as “Anthropic's own official SDK... for teams fully committed to Claude who don't need ADK's four-language, multi-provider reach,” and the Claude Agent SDK's rail reciprocally names ADK second. Both are official, first-party agent SDKs from a major AI lab, part of the same four-SDK category as OpenAI Agents SDK and Microsoft Agent Framework.

**Which is more open source?**

Google ADK, clearly. Its core is Apache 2.0-licensed and fully self-hostable for free across four production language SDKs. The Claude Agent SDK's Python and TypeScript wrapper packages are MIT-licensed, but Anthropic's own docs state overall SDK use is governed by its Commercial Terms of Service, and the compiled Claude Code CLI binary the wrapper bundles and spawns as a subprocess isn't published as source — the client is open, the engine underneath isn't.

**Can I use the Claude Agent SDK with Gemini or another non-Anthropic model?**

No. The Claude Agent SDK runs Claude models only (Opus 5, Sonnet 5, Haiku 4.5) — a GitHub issue asking Anthropic to support other providers was closed “not planned.” Google ADK is optimized for Gemini but is genuinely multi-provider through a LiteLLM integration, reaching OpenAI, Anthropic Claude, Cohere, local Ollama models and 100+ others.

**Which has more GitHub traction?**

Google ADK by a wide margin. Checked live via the GitHub API on 2026-08-07: google/adk-python carries 21,029 stars and 3,809 forks, versus a combined 9,504 stars across the Claude Agent SDK's two language repositories (7,826 on claude-agent-sdk-python, 1,678 on claude-agent-sdk-typescript). All three repos were pushed within the last day of this research.

**Do either support the Agent2Agent (A2A) protocol?**

Only Google ADK. It ships native support for both dominant 2026 agent-interoperability standards — A2A, now governed by the Linux Foundation with 150+ organizations reported in production, and MCP. The Claude Agent SDK supports MCP (both in-process SDK servers and external MCP servers) but has no A2A equivalent in its integrations list.