CONTRI

Know if a GitHub issue is really free before you start.

Claims, linked pull requests, maintainer reply times and project activity — right on GitHub.

Add to Chrome · Website · Features · For AI agents (MCP) · How the numbers work · Contributing

Why CONTRI?

Ever spent a weekend on a “good first issue”, only to find someone claimed it in the comments a month ago? Starting in open source is mostly guesswork: is the project still maintained, will anyone review my PR, and is someone already working on this? CONTRI answers those three questions on the page you're already on, using public evidence it links to — no backend, no analytics, public repositories only.

Related MCP server: contrib-radar

What it does

CONTRI follows GitHub's light or dark appearance with the same teal glass look.

Beyond the extension

All of it runs on public GitHub data with no CONTRI server: the lists are static files built by a scheduled workflow.

Install

From the Chrome Web Store (recommended)

Add CONTRI to Chrome — one click, updates automatically. The same listing works in Edge (allow extensions from other stores when asked) and Brave.

From a release

For Firefox (experimental) or a manual install:

1. Download contri-v<version>.zip (not “Source code”) from the latest release and unzip it.

2. Open chrome://extensions (or edge://extensions, brave://extensions) and turn on Developer mode.

3. Click Load unpacked and choose the unzipped folder.

4. A welcome page opens. Sign in with GitHub (or skip and use anonymous mode).

From source

Requires Node.js 22.12 or newer (CI and releases use 24, see .nvmrc).

npm ci
npm run build        # outputs dist/

Then Load unpacked → dist. After pulling changes, rebuild, click Reload on the extension card and reload your GitHub tab.

Firefox (experimental): npm run package:firefox creates contri-v<version>-firefox.zip for about:debugging → Load Temporary Add-on. Firefox treats the github.com permission as optional: after loading, open the add-on's Permissions tab and allow it, or nothing appears on GitHub. On-device AI is Chrome-only.

For AI coding agents (MCP)

AI agents increasingly open pull requests on issues someone already claimed. contri-mcp gives them CONTRI's checks as an MCP server, with the same reports and verdicts as the extension: check_issue, check_issues, repo_health, team_responsiveness, find_issues, discover_projects and contribution_brief. It runs locally over stdio, is read-only, talks only to api.github.com, and returns compact results (a verdict is about 40 tokens) so agents spend requests and tokens on the issue, not on scraping GitHub pages.

In Claude Code, install it as a plugin, which adds the server and two skills, /contri:can-i-take-this and /contri:find-first-issue:

/plugin install contri --marketplace alphasafal/contri

Or add the server alone:

claude mcp add --transport stdio contri -- npx -y contri-mcp

No token to paste: when CONTRI isn't signed in, the agent offers Sign in with GitHub, shows a short code, and you approve it on github.com, the same device flow as the extension. Setup for Cursor, Claude Desktop and other clients is in mcp/README.md; the plugin is in plugin/.

Signing in

Anonymous mode works within GitHub's limit of 60 requests per hour. Signing in raises it to 5,000 and switches CONTRI to GitHub's GraphQL API, so most checks take one request instead of dozens.

• Sign in with GitHub uses the OAuth device flow with the CONTRI OAuth App: you get a short code, approve it on github.com, done. No password ever passes through CONTRI, and no scopes are requested (public, read-only access).

• Personal access token: Settings walks you through creating a fine-grained, read-only token for public repositories.

The sign-in is kept in memory until the browser closes unless you choose “Stay signed in”, which stores it encrypted (AES-GCM). Either way it lives where GitHub pages cannot read it and is only ever sent to api.github.com. If GitHub rejects it (expired or revoked), Settings asks you to reconnect.

“Sign in with GitHub” in your own build

Every build uses the CONTRI OAuth App (client ID in src/shared/config.ts; client IDs are public and the device flow has no secret). A fork that wants its own app:

1. On GitHub: Settings → Developer settings → OAuth Apps → New OAuth App. Use your homepage as the callback URL (it is not used), tick Enable Device Flow, and leave Expire user access tokens off.

2. Copy the Client ID and build with VITE_GITHUB_CLIENT_ID=<your client id> npm run build (or put it in .env).

VITE_GITHUB_CLIENT_ID=none builds a version that offers only the token option.

Request budget

Reports are cached (team: 6 hours, overview: 30 minutes, issue checks: 10 minutes). Partial reports are cached for a third of that so large repositories don't burn requests on every visit.

Development

npm run dev          # layout preview of the widget (no live data)
npm test             # unit tests (fixtures, no network)
npm run test:e2e     # installs dist/ into Chromium, drives it on a mock GitHub, runs axe accessibility checks
npm run lint         # ESLint
npm run format       # Prettier
npm run check        # format check + lint + unit tests + build
npm run package      # dist → contri-v<version>.zip (+ sha256)
npm run verify:package
CONTRI_CONTRAST=1 npx playwright test contrast   # WCAG AA text contrast over the glass
CONTRI_PERF=1 npx playwright test perf           # content-script cost on GitHub pages

For test:e2e, run npm run build first and install the browser once with npx playwright install chromium.

The code is organized so contributors can work on one layer at a time:

src/core/        pure logic: metrics, maintainer roster, claim detection, verdicts (unit-tested)
src/background/  service worker: sign-in, request queue, caching, GitHub REST + GraphQL reports
src/content/     what runs on github.com: launcher, widget UI, badges, on-device AI
src/settings/    the Settings / welcome page
src/shared/      types and helpers shared by both sides
mcp/             the MCP server for AI agents: reuses src/core and the reports
action/          the GitHub Action for maintainers: same reports, bundle committed in action/dist
feed/            builds the daily data for the website's issue list, leaderboard and badges
website/dist/    the static website, including the data pages and Open Source Wrapped
tests/unit/      Vitest
tests/e2e/       Playwright + mock GitHub

More detail in docs/ARCHITECTURE.md.

New to the codebase? DeepWiki has a browsable walkthrough of every part of CONTRI — the service worker, claim detection and verdicts, metrics and the maintainer roster, the widget, tests and release pipeline — and you can ask it questions about the code. It is generated automatically from this repository, so if it disagrees with the code or docs/, the code and docs/ win.

Honest numbers

Every number links to its evidence. Unavailable data is shown as “—”, never as zero. Samples say how much they cover, and ≥ marks a lower bound. A missing linked PR does not prove nobody is working on an issue, and CONTRI says so. Full definitions: docs/METRICS.md.

Known limitations

• Public evidence only. CONTRI sees what GitHub's public API shows. Work nobody linked or mentioned is invisible, so “Looks free” is a strong hint, not a guarantee.

• Samples on big projects. Charts, author counts and team reply times use the most recent items (up to 100 merged PRs, 20 PRs and issues per kind, the latest 100–300 comments of an issue). Every sampled number says so, and lower bounds are marked ≥.

• Claim detection is a heuristic over English text: a first-person offer (“I'd like to”, “can I”, “I'll”), an action (“work on”, “take”, “fix”) and a target (“this”, “the issue”, “a PR”), plus /assign-style commands. Other languages and unusual wording may be missed; report a wrong verdict and it becomes a test.

• Issue list badges need sign-in. One page of issues is one GraphQL request when signed in, but about 75 REST requests anonymously, more than GitHub's hourly limit. They read the latest 20 comments per issue; the issue page reads up to 100.

• Maintainer detection uses GitHub's team badges, triage and merge activity, CODEOWNERS and the owner. Maintainers who never triage or merge in the sampled window may be missed.

• GitHub layout changes can move or hide the inline badges until CONTRI is updated. The panel keeps working.

• On-device AI is Chrome-only for now. Summaries need a desktop Chrome with the built-in model (Gemini Nano). Everything else — badges, verdicts, team reply times, discovery — uses no AI and works in every supported browser. Without a built-in model CONTRI shows the latest comments and offers the copy-paste brief.

• Anonymous mode shares GitHub's limit of 60 requests per hour with anything else on your network.

Troubleshooting

Contributing

Issues and pull requests are welcome — especially for accessibility, metric accuracy, GitHub layout changes that break badges, and translations. Start with CONTRIBUTING.md. Good first issues are labeled good first issue (CONTRI will tell you whether they're free).

FAQ

Only to GitHub's API, to read public repository data. There is no CONTRI server and no analytics. On-device AI runs inside Chrome. See PRIVACY.md.

No. It means CONTRI found no assignee, no recent claim and no linked open pull request. Someone may be working without saying so — always comment before you start.

Timings need at least three answered items in the sample. Small or quiet projects often don't have that many recent outside contributions, and CONTRI won't guess.

No, by design. CONTRI is for joining public open-source projects.

Chrome, Edge, Brave and other Chromium browsers (version 124+). Firefox 128+ is experimental. On-device AI needs a Chrome build with built-in AI.

No. Verdicts, claims, linked pull requests, reply times and activity are plain rules over public GitHub data, documented in docs/METRICS.md, and work the same in every supported browser. AI is only used for two optional buttons (“Summarize the thread”, “How do I start?”), and only the model built into your browser, on your device. Where the browser has none, CONTRI shows the latest comments as written.

Roadmap

• Issue availability badges with claim detection

• Maintainer roster beyond GitHub's team badge

• Sign in with GitHub (device flow) and GraphQL reports

• On-device AI summaries

• Watched issues with notifications

• Chrome Web Store listing

• MCP server, so AI coding agents check an issue is free before they start

• Badges on a repository's issue list (signed in)

• Maintainer “I'm on it” statements as their own verdict

• GitHub Action for maintainers: duplicate-work check and quiet-claim nudges

• “Free right now” issue list, reply-time leaderboard and README badges

• Open Source Wrapped

• Badges on GitHub's global issue search results

• Translations

• Maintainer view: who is waiting longest for a reply

Have an idea? Open a feature request. Thanks to everyone who has contributed.

If CONTRI helped you find your first contribution, a ⭐ helps other newcomers find it too.

License

Built and maintained by Safal Gupta (@alphasafal).

MIT © Safal Gupta. See LICENSE and THIRD_PARTY_NOTICES.md.