# Ownhand > Ownhand is an MCP server that writes in the user's own voice, fitted to the occasion, and learns from the text the user actually sends. Connect any MCP client to https://ownhand.dev/mcp (streamable HTTP). The user signs in through the browser with OAuth, so no API key is needed. Each user has a Hand: a voice profile built from 5 to 10 things they wrote themselves. New accounts get $1 of free credit, then pay as you go per token. The core loop: call `write` with the draft and an occasion, show the user the draft as returned, and never send anything without their OK. After the user sends it, call `send_feedback` with the `request_id` and the exact text that was sent. That is how the Hand learns. There are 12 occasions. Pick one from where the text will be posted and who will read it: `chat_dm`, `chat_channel`, `email_internal`, `email_formal`, `email_cold`, `email_warm`, `proposal_cold`, `proposal_warm`, `pr_description`, `review_comment`, `docs`, `status_update`. ## Setup - [Setup prompt](https://ownhand.dev/agent-setup/prompt.md): fetch and follow this to install Ownhand in any MCP client. - [MCP endpoint](https://ownhand.dev/mcp): streamable HTTP with OAuth sign-in. No API key needed. - [Connect](https://ownhand.dev/connect.md): commands and configs for Claude Code, Claude, ChatGPT, Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed, Goose and Cline. ## Docs - [Docs](https://ownhand.dev/docs.md): quickstart, occasions, replies, learning from feedback, the HTTP API, and errors. - [Agent guide](https://ownhand.dev/docs/agent-guide.md): the playbook for agents using the tools, with the occasion table and what to do on each error. - [Pricing](https://ownhand.dev/pricing.md): per-token prices, the free credit, and reloads. ## Answers - [All answers](https://ownhand.dev/answers.md): short answers about making AI agents write in your own voice. - [How do I make Claude write like me?](https://ownhand.dev/answers/make-claude-write-like-me.md): Every way to get Claude to write in your voice: instructions, Projects with samples, skills, memory, Claude Code, and Ownhand. Steps and limits for each. - [How do I make ChatGPT write in my voice?](https://ownhand.dev/answers/make-chatgpt-write-in-my-voice.md): Every way to get ChatGPT to write like you: custom instructions, Projects with samples, memory, base style, GPTs, and Ownhand. Steps and limits for each. - [How do I stop AI emails and Slack messages sounding like AI?](https://ownhand.dev/answers/stop-ai-drafts-sounding-like-ai.md): Free fixes in ChatGPT, Claude, Cursor and Copilot, when they are enough, and how a voice layer rewrites each draft from your own samples. - [Which phrases make a message read as AI-written?](https://ownhand.dev/answers/ai-writing-tells.md): The openers, words and sentence shapes that make a message read as AI-written, grouped, with real examples and a plain swap for each. - [Why do AI drafts sound like AI, and what fixes it?](https://ownhand.dev/answers/why-ai-drafts-sound-like-ai.md): Six causes, from how chat models are tuned to the thread they never see, with the research behind each, real before and after pairs, and a fix for each. - [How do I write a cold email with AI that doesn't read as AI?](https://ownhand.dev/answers/cold-email-with-ai-that-doesnt-read-as-ai.md): Keep it to 50 to 125 words: why you, why them, one small ask, and a way to say no. Real AI drafts and rewrites show the difference. - [Is there an MCP server that writes in my voice?](https://ownhand.dev/answers/mcp-server-for-writing-in-my-voice.md): Yes: Ownhand and Idiolect are MCP servers built from your own writing samples. How they differ, how to connect one, and when you need neither. - [How do I get Claude Code or Cursor to write PR descriptions in my style?](https://ownhand.dev/answers/pr-descriptions-in-my-style.md): Four ways to get Claude Code, Cursor or Copilot to write pull request descriptions like you, from a free rules file with examples to an MCP server. - [How do I teach an AI agent my writing style?](https://ownhand.dev/answers/teach-ai-my-writing-style.md): Real samples beat a description, and the agent needs to see what you send. Free options in Claude, ChatGPT, Cursor and Copilot, and where Ownhand fits. - [Ownhand or Idiolect: which should I use?](https://ownhand.dev/answers/ownhand-vs-idiolect.md): An honest comparison of Ownhand and Idiolect, two MCP connectors that make AI write in your voice, checked against each one's public pages. - [Should I use Ownhand or a style prompt in custom instructions?](https://ownhand.dev/answers/ownhand-vs-custom-instructions.md): A style prompt is free and fixed. Ownhand writes from your own samples, per occasion, and learns from what you send. When each one fits. - [How does Ownhand learn from my edits?](https://ownhand.dev/answers/how-ownhand-learns-from-edits.md): After you send a draft, your agent reports the exact text with send_feedback. Reasons become rules at once; patterns in edits after two. - [Is Ownhand safe to use, and what does it store?](https://ownhand.dev/answers/is-ownhand-safe.md): What Ownhand can do in your agent, what it stores and never stores, who processes your text, the one training trade-off, and how to delete it. ## Benchmarks - [Benchmarks](https://ownhand.dev/benchmarks.md): how Ownhand's rewrites compare with frontier models and commercial humanizers on slop, meaning kept, facts kept, length and voice match, with intervals. - [How we tested](https://ownhand.dev/benchmarks/methodology.md): the test set, the systems, how each score is computed, and the caveats. - [Benchmark data](https://ownhand.dev/benchmarks.json): every number on the benchmarks page, as JSON. ## Optional - [Home](https://ownhand.dev/index.md): what Ownhand is, in one screen. - [Full text](https://ownhand.dev/llms-full.txt): this file with the docs, connect, pricing, agent guide and setup prompt inlined. - [Terms](https://ownhand.dev/terms.md): the terms of service. - [Privacy](https://ownhand.dev/privacy.md): what is stored and who processes it. - [Support](https://ownhand.dev/support.md): who to email, what to include, and where to check service status.