OpenDocs for Chrome and Edge
Publish any ChatGPT, Claude or Gemini answer, or a whole chat, as a clean
OpenDocs document: a link you can share and a PDF. No
account needed to try it; sign in to keep documents in your workspace, include
images and update them at the same link.
Install it from opendocs.cc/extension.
How it works
- A Publish button sits in each answer's action row. It opens a card with
the title, whether to include your question, where the document goes and
Publish. Switch the card to Whole chat to publish the conversation,
with each question quoted above its answer.
- The toolbar icon opens a side panel: your account, where documents go
(workspace, project, who can see them), recent publishes, and fallback
buttons that work even if a site's redesign hides the in-page buttons.
- Without an account, documents are anonymous: anyone with the link can read
them for 24 hours. Signing in moves this browser's anonymous documents into
your workspace.
Reading chats
Each site has a reader in src/sites/<site>/:
| Site | Main source | Fallback |
|---|
| ChatGPT | The conversation data the site itself loads (/backend-api/conversation/<id>): the model's own Markdown, the branch on screen, citations, canvases | The rendered page |
| Claude | /api/organizations/<org>/chat_conversations/<id>: Markdown blocks and artifacts at their final version | The rendered page |
| Gemini | The rendered page, with math from data-math | |
Readers run in the page, with the site's session, and only after a click.
src/markdown/ turns what they read into the document: LaTeX to $…$,
citations to numbered footnotes, images to uploads (signed in) or a note
(signed out), questions as quotes.
When a reader fails, the extension sends OpenDocs a failure ping with the
site, the step and an error code, never content
(POST /api/v1/extension/events), so breakage after a site redesign shows up
quickly.
Talking to OpenDocs
The service worker (entrypoints/background.ts) is the only part that holds
tokens or calls OpenDocs. It signs in with OAuth 2.1 and PKCE through
chrome.identity.launchWebAuthFlow, using the client metadata document at
https://opendocs.cc/oauth/clients/browser-extension.json, and calls
api.opendocs.cc, which answers the extension's origin with CORS. The
extension has no host permission for OpenDocs, so the install dialog names
only the three chat sites.
Permissions
| Permission | Why |
|---|
| Content scripts on chatgpt.com, claude.ai, gemini.google.com | Show the Publish button and read the chat you publish |
identity | The sign-in window |
storage | Settings, sign-in tokens, recent publishes |
sidePanel | The side panel |
Development
Requires Node.js 20+.
npm install
npm run dev # Chrome with hot reload
npm run build # .output/chrome-mv3, load it unpacked in chrome://extensions
npm test # unit tests (Vitest)
npm run test:e2e # end-to-end (Playwright: builds the extension, runs it in Chromium)
npm run typecheck
Development builds carry the key in keys/dev.pub, so the unpacked
extension always gets the ID pmpocloilkjljlpelbkngppeaagmaccc, which
OpenDocs knows (sign-in redirect URL and CORS).
Against a local OpenDocs stack
Run the OpenDocs web app and API locally, register a sign-in client (Better
Auth only accepts https client metadata URLs, so locally it's a registered
one):
curl -X POST http://localhost:3150/api/auth/oauth2/register -H 'Content-Type: application/json' \
-d '{"client_name":"OpenDocs extension (local)","redirect_uris":["https://pmpocloilkjljlpelbkngppeaagmaccc.chromiumapp.org/"],"token_endpoint_auth_method":"none","grant_types":["authorization_code","refresh_token"],"response_types":["code"],"scope":"docs:read docs:write offline_access"}'
Copy .env.example to .env.local with that client_id, then npm run dev.
tests/e2e/local-stack.spec.ts runs the whole flow (anonymous publish, sign
in through the consent page, claim, workspace publish, PDF) against such a
stack.
Releasing
npm run release:chrome # .output/opendocs-extension-<version>-chrome.zip
npm run release:edge # .output/opendocs-extension-<version>-edge.zip
Store packages leave out the development key; each store assigns its own
extension ID. When a store ID is new, add it to EXTENSION_IDS on the OpenDocs
servers (sign-in redirect and CORS), or to DEFAULT_EXTENSION_IDS in
packages/shared/src/extension/config.ts. After the first Chrome Web Store
upload, put the store item's public key in keys/dev.pub so development builds
share the store ID.
store/ has the listing text, the privacy answers for each store and the
screenshots (SCREENSHOTS=1 npx playwright test screenshots).
License
MIT
