Your agent can already read the repo. It can't remember how you write.

RepoMuse keeps your author profile and your real past posts as versioned state, then drafts for X, Threads, LinkedIn and Bluesky. Call it from Claude, Cursor, or the web. It never publishes.

YOUR AGENT

Your agent writes. We hold the state.

Call RepoMuse from Claude, Cursor, or any MCP client. Your model drafts locally. We supply the voice corpus, repo research, and platform validation. That loop is free.

X Threads LinkedIn

YOUR VOICE

Your voice is a record, not a prompt

Voice comes from the author profile you define and your own past posts, kept as versioned state. Anything scraped from the web is context only and never becomes voice evidence.

your repo your author profile your past posts drafting

YOUR REPO

Grounded in what you actually shipped

Optionally link a read-only repository and RepoMuse reads the README, releases, merged PRs, and commits. Repo content is loaded first, so what shipped outranks a marketing page.

README
merged PRs
releases

YOUR CALL

The last edit is always yours

A finished draft for X, Threads, LinkedIn, or Bluesky. RepoMuse never publishes, so nothing goes out until you send it.

X Threads LinkedIn

What you get

  • Free MCP path: your agent writes locally. RepoMuse supplies the voice corpus, repo research, and platform validation.

  • A chat model can read your repo. It cannot read the forty posts you already wrote - RepoMuse keeps them as a versioned corpus and drafts against them.

  • Link a private repo. We store no user credential: only a numeric GitHub installation ID. Every call mints a ~1 hour read-only token.

  • Writes for X, Threads, LinkedIn, and Bluesky, with per-platform length rules enforced before you see the draft.

  • Free is 15 posts a month. Pro is $12/mo for 300.

From the founder

The compose box was never the problem. I had plenty to post. The problem was sitting there turning one idea over and over, angle, hook, wording, second-guessing every draft, until 30 minutes were gone.

Then the tweet went out and nobody read it. Same generic take everyone else already posted that day, just slower to write.

So instead of trying to think faster, I built something that already knows the material: it reads the repo I just pushed to and the posts I have already written, then drafts from those instead of from a blank box.

That's RepoMuse. Still early. If you've ever burned 30 minutes on one tweet nobody read, give it a try.

FAQ

How is this different from asking ChatGPT or Claude directly?

A chat model can read your repo. It cannot read the forty posts you already wrote. A CLAUDE.md remembers instructions; RepoMuse keeps your author profile and real past posts as versioned state and drafts against them.

Is there an API?

Yes, MCP, and the free path is built for it. Your agent fetches a drafting packet, writes locally, validates against platform rules, and stores a draft. Same voice corpus as the web studio. It never publishes.

How is this different from Typefully, Hypefury, or Postwise?

Those schedule and publish what you have already written. RepoMuse writes the draft and never publishes. Different job, and you can use both.

What does it actually read before writing?

Your linked repository first: README, releases, merged PRs, and commits. Then pages from your own site, and a web and Hacker News search on your product and category. It does not read anyone's social feed, including your competitors'.

What platforms does it write for?

Anything mostly text: X, Threads, LinkedIn, Bluesky. If it needs a caption or a post, it works.

How much can I generate?

Free includes 15 posts a month. Pro is $12/mo for 300. Extra packs never expire. Your agent can draft for free over MCP.

Coming from Typefully or Hypefury? See the comparison. Building in public? Read the guide, or the developer-focused writeup. Wiring up an agent? See MCP access.

Stop starting from a blank prompt.