Is there an MCP server that writes in my voice?

Yes. Two MCP servers are built for this. Ownhand is a remote server at https://ownhand.dev/mcp. It rewrites drafts from a profile of 5 to 10 of your samples, fits them to 12 occasions, and learns from the text you send. Idiolect builds a profile from 3 to 5 examples or a short interview. It's free for personal use during its beta. If you only write in one chat app, samples in Claude project instructions or ChatGPT custom instructions may be enough without any server.

By Jordan Gibbs, founder of Ownhand at Thalient Labs. Updated 2 October 2026.

An MCP server is the right shape for this job. MCP is the open standard that lets apps like Claude, ChatGPT, Cursor and VS Code call outside tools. A voice server keeps your samples and rules in one place, and every client you connect to calls the same tool before showing you a draft. A custom instruction lives in one app and only describes you. A server goes with you.

The options

Option	Built from	Works in	Learns from what you send	Price
Ownhand	5 to 10 things you wrote	Any remote MCP client	Yes, from the exact text you send	$1 free, then about half a cent a rewrite
Idiolect	3 to 5 examples, or a short interview	Claude, ChatGPT and other MCP clients	A learn_writing_style tool is listed; learning from sent text is not described	Personal plan free during beta
Claude instructions, projects or skills	Whatever you paste in	Claude only	No	Included in your plan
ChatGPT custom instructions or projects	Whatever you paste in	ChatGPT only	No	Included in your plan
A server you build	Your own prompt and samples	Any MCP client	Only if you build it	Your model costs and your time

Ownhand

Ownhand is a remote server over streamable HTTP. Add the URL, sign in through the browser with OAuth, pick which Hand the connection uses, and click Allow. It's listed in the official MCP Registry as io.github.jordan-gibbs/ownhand, and the install configs for each client are in the public repo.

Tool	What it does
`write`	Rewrites a draft in your Hand for an occasion. For replies, takes up to 30 messages of the thread as `context_messages`.
`send_feedback`	Reports what you did with a draft and the exact text you sent. Your Hand learns from it.
`get_hand`	Lists your Hands, or shows what one has learned.
`create_hand`	Builds a Hand from 5 to 10 things you wrote.
`update_hand`	Adds samples, sets how you describe your writing, or pins, rejects or activates a learned rule.
`get_account`	Shows your account, credit and usage.

The occasions are chat_dm, chat_channel, email_internal, email_formal, email_cold, email_warm, proposal_cold, proposal_warm, pr_description, review_comment, docs, status_update. Your agent picks one based on where the text is going and who reads it. Each has its own rules.

A chat_dm stays under 60 words with no Markdown. An email_cold stays under 125 words and has a greeting and a sign-off. A pr_description has neither, but can use Markdown headings.

Each draft is checked before you see it. The checks look for listed AI words and sentence patterns, the occasion's rules, and facts. Every name, number, link, and email address in your draft has to still be there. If a check fails, the model gets the list and tries again, up to two more times. Anything left is reported to your agent as unresolved.

Replies match the thread. Your agent passes the recent messages verbatim. Ownhand measures how you write in that thread and matches it. The thread is used for that one request and never stored.

Here is the same AI draft rewritten by two demo writers, then a status update by a third:

These examples are real output from the live API. Each AI draft was rewritten by a demo Hand: a fictional writer (an engineer, a founder, an ops lead, a designer or a consultant) built from a handful of samples.

AI draft: Hi Priya! I hope you're doing well. I wanted to reach out to see if you might have a moment to review my PR #482 before the 3pm deploy.

Demo Hand, engineer: priya, can you review pr #482 before the 3pm deploy?

AI draft: Hi Priya! I hope you're doing well. I wanted to reach out to see if you might have a moment to review my PR #482 before the 3pm deploy.

Demo Hand, designer: Hey Priya, could you review PR #482 before the 3pm deploy?

AI draft: I'm pleased to report that the migration is approximately 80% complete. We remain on track to finish by Friday, barring any unforeseen issues.

Demo Hand, founder: The migration is about 80% done. We're on track to finish by Friday, unless something comes up.

The engineer writes in lowercase and cuts the greeting. The designer keeps a "Hey" and a capital P. The facts stay the same. The people don't. That difference comes from the samples. An instruction like "casual and short" can't produce it.

Then it learns. After you send a draft, your agent calls send_feedback with the exact text that went out. A reason you give becomes a rule at once. A pattern in your edits becomes a rule after it shows up twice. See how Ownhand learns from what you send.

Connect it

Claude Code: run the command under Install, then /mcp, pick ownhand and sign in.
Claude on the web or desktop: open Customize, then Connectors, click the plus button and Add custom connector, and paste https://ownhand.dev/mcp. Free plans can add one custom connector.
ChatGPT: open Settings, then Security and login, and turn on Developer mode. OpenAI lists it for Pro, Plus, Business, Enterprise and Education accounts on the web. Then add an app for the server URL with the plus button under Plugins.
Cursor: use the button on Connect, or add ownhand with its URL to ~/.cursor/mcp.json.
Then say "set up my Ownhand Hand" and paste 5 to 10 things you wrote: Slack messages, emails, PR descriptions. Only things you sent yourself. AI drafts make a poor Hand.

To make it automatic, paste the agent blurb from the docs into CLAUDE.md, AGENTS.md or your custom instructions. Agents can install it themselves from the setup prompt. /llms.txt is the index for agents.

Idiolect

Idiolect is the other server built from your samples. Its remote MCP endpoint is idiolect.app/api/mcp, with OAuth 2.1 sign-in from claude.ai or ChatGPT. The developer page lists five tools: write, rewrite, refine_writing, manage_profile, and learn_writing_style. The Score API rates how close a draft is to your voice from 0 to 100, using an authorship model called LUAR-MUD. There's also a web studio for drafting. The personal plan is free while it's in beta.

Pick Idiolect if you want to pay nothing for now, want a score for voice match, or want to draft in a web app. Pick Ownhand if your writing changes by occasion, you reply to threads a lot, or you want your voice to keep improving from what you send. See Ownhand vs Idiolect for the full comparison.

Searching the MCP Registry for writing tools also turns up general tone rewriters. Most apply a named tone like formal or friendly. That won't make a draft sound like you. Before you connect one, check if it builds a profile from your own samples.

When you do not need a server

If you write in one app, start with what it already has. Anthropic's prompting guide suggests 3 to 5 examples for steering tone and structure. Paste in real messages you wrote. A description of your style does less.

Claude: add account-wide instructions in Settings. Or make a project, free accounts get up to five. Put your samples in the project instructions. Claude skills can also carry your writing patterns.
ChatGPT: open Settings, then Personalization, and use Custom Instructions. Free and Go plans get 1,500 characters. Paid plans get 5,000, which fits a few short samples. A project holds more. It has its own instructions plus files, 5 per project on Free and 25 on Plus.

This is enough when you write in one app, a few times a week, for one kind of reader. It falls short when you write in several tools, when a Slack reply and a client email need different registers, or when you're tired of fixing the same habit by hand. A server fills those gaps. For more on the free route, see Ownhand vs custom instructions. Before you connect any server, read whether Ownhand is safe.

Follow-up questions

Do I need to run anything to use an MCP server?

Not for a remote one. Paste the URL into your client and sign in through the browser. Ownhand has nothing to install and no API key to copy. A key is only needed for scripts, CI and clients that can't sign in.

Which apps can use Ownhand?

Any client that supports remote MCP servers works. The Connect page has steps for Claude Code, Claude on the web, desktop and mobile, ChatGPT, Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed, Goose and Cline. Clients that only run local servers can bridge with mcp-remote.

Is my writing used to train AI models?

Ownhand doesn't train its own models on your data. Learning from feedback runs on a model through OpenRouter's contributor tier. That provider may use what it receives to improve its models, including your samples when your style card is rebuilt. Drafts you only rewrite don't go to it. The privacy policy lists exactly what's sent.

Can I use it without MCP?

Yes. There's a plain HTTP API that takes the same draft, Hand and occasion. You create an API key in the dashboard. It works for scripts, CI jobs and agents that don't speak MCP. The docs have a curl example.

How is this different from asking the model to write like me?

A request in chat gives the model a description of you. It fills the gaps with its defaults. Ownhand keeps your samples and learned rules on the server. It applies them on every call from every client and checks each draft before you see it.

Sources

Checked on 2 October 2026. If something here is out of date, email hi@thalientlabs.ai and we will fix it.

Model Context Protocol: What is MCP?: the open standard for connecting AI apps to tools
Official MCP Registry entry for io.github.jordan-gibbs/ownhand: streamable HTTP remote at ownhand.dev/mcp
Claude Help Center: Getting started with custom connectors using remote MCP: Customize, Connectors, Add custom connector; Free plans get one custom connector
OpenAI docs: Developer mode: Pro, Plus, Business, Enterprise and Education on the web; Settings, Security and login
Claude Code docs: MCP: claude mcp add, scopes, and signing in with /mcp
Cursor docs: MCP: ~/.cursor/mcp.json and remote URLs
Claude Help Center: Understanding Claude's personalization features: profile instructions, project instructions, skills; five projects on free accounts
OpenAI Help Center: ChatGPT custom instructions: Settings, Personalization; 1,500 characters on Free and Go, 5,000 on paid plans
OpenAI Help Center: Projects in ChatGPT: project instructions and files, file limits by plan
Claude docs: Prompting best practices: 3 to 5 examples for best results
idiolect.app/developers: remote MCP at idiolect.app/api/mcp, OAuth 2.1, its tools, the Score API
github.com/moshpit-labs/idiolect-mcp: profile from 3 to 5 examples or a guided interview
idiolect.app/pricing: personal plan free during the beta
Ownhand public repo: install configs for each client

Install

In Claude Code it is one line. You sign in through your browser, so there is no API key to copy.

For Claude, ChatGPT, Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed, Goose and Cline, see Connect. New accounts get $1 of free credit. After that you pay as you go, and each rewrite costs about half a cent.