Built-in agent
Ready the moment you sign up. Fast, thoughtful, and native to every ThinkBench surface — no setup required.
One bench. Any agent.
ThinkBench ships with a powerful agent that's ready the moment you sign up. Rather bring your own? Claude Code, Codex, and anything else can pull up a seat through our MCP server, CLI, and SDK.
Start free
No card required.
MCP server
The default bridge for Claude Code, Codex, and any other MCP-speaking agent. One command and it's connected.
CLI tools
The same capabilities as plain commands — a natural fit for scripts, pipelines, and terminal-first workflows.
SDK
Typed building blocks for custom agents and deep integrations, straight from your own codebase.
01
No model lock-in
Use any model
Pick the right model for the task — not the one you're locked into.
Every agent in ThinkBench runs on the model you choose. Claude, Gemini, OpenAI, DeepSeek, Qwen, Kimi, and more — all available from one picker, with new models arriving as they ship.
Switch whenever it makes sense: a fast model for cleanup passes, a reasoning model for the section that actually matters. Your choice applies instantly, mid-document, with nothing to reconfigure.
02
MCP server
Start with MCP
The default way in for Claude Code, Codex, and any MCP-speaking agent.
If your agent speaks MCP, you're one command away: add the ThinkBench server and it can read documents, propose focused edits, and leave comments through the same interfaces our built-in agent uses.
It's the right choice for most people. Coding agents support it out of the box, and everything your agent does still lands as visible suggestions and comments you review — never silent rewrites.
03
CLI tools
Or drive from the terminal
Everything the MCP server offers, as plain composable commands.
Prefer commands over a protocol? The CLI covers the same surface: list documents, pull content, push suggestions, leave comments — each one a small command that pipes like everything else in your shell.
That makes it the natural alternative for scripts and automation. Wire ThinkBench into a cron job or CI pipeline, or hand it to an agent that's happier running shell commands than speaking MCP.
04
SDK
Go deeper with the SDK
For when an integration deserves real code.
Building something more ambitious than a one-off command? The SDK gives you typed access to documents, suggestions, comments, and reviews from your own codebase.
It's the advanced path: custom agents, internal tools, and integrations that need full control over how work flows in and out of ThinkBench.
Documentation
View the docs Ready to wire one up?
MCP setup guides, the full CLI reference, and the SDK API — everything an agent needs to get to work.
Put an agent to work.
Start with ours, connect your own whenever you like. The bench is open.
Start free