Rill

Docs

Writing from Claude Code

New to this? Copy the prompt and paste it into Claude Code — it walks you through the whole setup and tells you what to put where.

Write notes from the session where the work happened, without leaving it. There is no API key to create and nothing to paste — the connection authenticates with your Rill account.

Connect it

claude mcp add --transport http rill https://rill.page/api/mcp

The first time a tool runs, your browser opens and asks whether to let Claude Code write to your workspace. Approve it once and that is the whole setup.

Nothing is stored on your side but a token that expires and refreshes on its own. Revoke it whenever you like, per device, without touching your API keys or affecting your website.

Then just ask

"Write this up as a note."

Notes are created as drafts by default, so you can edit one in Rill before anything goes live. Say "publish it" to skip that.

The tools

rill_write_note           Create a post, with provenance attached
rill_update_post          Edit an existing post
rill_find_similar         Check whether you have written this already
rill_list_posts           Find something already written
rill_get_post             Read a post back as markdown
rill_start_image_upload   Get a URL to upload an image to
rill_finish_image_upload  Confirm it arrived, get markdown to embed
rill_upload_image         Small images, sent inline
rill_workspaces           List the workspaces you can write to

It writes in your voice

How you sound, what you write about, and where to look for material live in a skill in your own repository — one editable file you own, written during setup by answering a few questions. The setup prompt on the Getting started page installs it. Change any line of it whenever you like; nothing about your voice is hidden on a server.

To match your voice, the model also reads a couple of your recent posts before writing.

It checks before it duplicates

Before writing, the model can search what you have already published for anything close in meaning. Ask for a note on something you have covered before and it will find the earlier post and offer to update it, rather than quietly writing a second one. Matching is by meaning, not keywords — a note about "flaky tests" finds one titled "making CI deterministic".

Screenshots

Ask in passing — "add the screenshot from the failing test" — and the image is uploaded and referenced in the post.

Rill hands out a signed URL and the file goes from your disk straight to storage. It never travels through the conversation, so a full-resolution screenshot costs nothing and does not need cropping or recompressing to fit.

Provenance

A note can carry the commit and the changed files it came from, so it links back to the actual code. Claude Code reads that from your repository and sends it along.

Rill never touches your machine. It runs on a server and sees only what your client chooses to send, which means file access stays under the permissions you already control in Claude Code rather than being handed to a third party.

More than one workspace

If you belong to several, the tools ask which one rather than guessing, and list your options. Say "write it to the acme workspace" and it goes where you meant.

Other clients

Any MCP client supporting remote servers and OAuth works the same way. Point it at:

https://rill.page/api/mcp

It discovers where to authenticate on its own — there is nothing to configure beyond the URL.