AIRANKS — The Authoritative Rankings for AI Web Content

AIRANKS measures AI visibility: we ask AI models real product and service questions, capture the complete answers as immutable observations, and publish what they contain — which brands were mentioned, which domains were cited, and which exact pages were linked. Every domain gets an AIR score from 1–10 (a decile of visibility in the active dataset; 0 means insufficient data), with the methodology in the open.

AIR

OPEN SOURCE

Skip to main content

contributions to the tools AI runs on

Open Source

AIRANKS is proud to give back to the open source projects it depends on. Measuring AI is not the only work AIRANKS does. Founder Jeremy Schoemaker works hands-on inside the same tooling that frontier labs and standards bodies ship: the Model Context Protocol SDK, the OpenAI Agents SDK, Claude Code's GitHub Action, and the CLIs and MCP servers Mozilla AI, Mozilla, and MiniMax maintain, and sends fixes back upstream when something is broken. Every pull request below merged to the project's main branch; each one links straight to it.

Nothing below is inferred. Every count on this page is computed from the list of pull requests it renders; the sample is the page.

The same care is how we measure you: every AIR ranking is a 1,000-resample bootstrap with a visible confidence interval, not a confident single number. Run your free report to see yours -

the ledger, computed from the list below

merged pull requests
16
organisations
7
repositories
9

Provenance: PR merge dates 2026-08-03 → 2026-08-24, verified via gh api repos/{owner}/{repo}/pulls/{n} against the live GitHub API on 2026-08-24; every PR above links to its merged commit. Full source in resources/js/data/contributions.js and loop-5 HAR at council_meetings/free-air-report-loop5-20260824-1236/verdict.html.

Each mark below identifies the organisation whose repository received a merged pull request; it is not a claim of sponsorship, partnership, or endorsement.

OpenAI

Headline merge1 PR
openai/openai-agents-python ↗OpenAI's official Agents SDK

Model Context Protocol

2 PRs
modelcontextprotocol/rust-sdk ↗the official Rust SDK for MCP
  • The SSE event loop skipped filtered events by recursing instead of looping. A long enough run of skips could overflow the stack and abort the client process outright.

  • The standalone SSE stream handled a 405 response but not a 401/403 carrying a WWW-Authenticate challenge, so an expired token there was never routed to the client's refresh-and-retry logic; the stream just failed. Mapped it the same way the sibling code path already did.

Anthropic

1 PR
anthropics/claude-code-action ↗the official Claude Code GitHub Action
  • The Bedrock section link in the docs pointed at an anchor that no longer exists, dropping readers at the top of a long page instead of the section they wanted. Repointed it at the anchor the page actually uses.

Mozilla AI

8 PRs
mozilla-ai/mcpd ↗a daemon for running MCP servers
mozilla-ai/apron-tools ↗MCP tools for Google Workspace
  • Gmail's API returns message bodies as base64url that commonly omits its trailing padding, which the decoder rejected silently, inside a caught exception, turning a readable email into placeholder text on roughly two-thirds of message lengths. Restored the padding before decoding, in one shared helper used everywhere the module decodes this format.

Mozilla

1 PR
mozilla/firefox-devtools-mcp ↗Firefox DevTools' MCP server

MiniMax

2 PRs
MiniMax-AI/cli ↗the official MiniMax CLI
MiniMax-AI/MiniMax-Provider-Verifier ↗a conformance verifier for MiniMax API providers
  • Streaming requests converted a failure into a normal return value instead of raising it, so the retry loop never saw the failure and gave up after one attempt; while non-streaming requests on the same provider retried up to ten times. That made a flaky provider score a better success rate under streaming than without it. Re-raised the failure so both modes retry the same way.

TanStack

1 PR
TanStack/ai ↗a type-safe, provider-agnostic TypeScript AI SDK