@metyatech/ai-quota

AI agent quota/rate-limit fetching library for Claude, Gemini, Copilot, and Codex.

This package extracts the quota fetching layer from agent-runner so it can be reused independently. Gate/ramp evaluation logic (e.g. evaluateUsageGate) is intentionally kept out of this package — it remains in the calling application.

CLI

After installing the package globally or via npx, run ai-quota to check quota for all agents at once.

# Install globally
npm install -g @metyatech/ai-quota

# Or use with npx (no install required)
npx @metyatech/ai-quota

Commands

ai-quota [agent]           Show quota for all agents, or a single named agent
ai-quota --json            Machine-readable JSON output
ai-quota --mcp             Start as an MCP server
ai-quota --quiet           Suppress non-error output (useful in scripts)
ai-quota --strict          Exit non-zero if any provider fetch returns error
ai-quota --verbose         Print debug info to stderr
ai-quota --help            Show usage information
ai-quota --version         Show version

Supported agent names: claude, gemini, copilot, codex

Usage Examples

Check all quotas:

ai-quota

Model Context Protocol (MCP)

ai-quota can act as an MCP server, allowing AI agents (like Claude Desktop) to check your remaining quota automatically.

Setup for Claude Desktop

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "ai-quota": {
      "command": "npx",
      "args": ["-y", "@metyatech/ai-quota", "--mcp"]
    }
  }
}

This exposes the get_quota tool to Claude, which it can use to stay aware of its own usage limits across different models and providers.

MCP Resources

ai-quota also provides an MCP resource:

• URI: quota://current
• Description: A live, auto-updating Markdown table of all current AI agent quotas.

AI agents can "subscribe" to this resource to keep the quota information in their context without needing to explicitly call a tool.

Tool: get_quota

• agent (optional): Specific agent to check (claude, gemini, etc.). If omitted, returns quota for all agents in a Markdown table.

Human-readable output example

AGENT        STATUS     LIMIT  DETAILS
-----------  ---------  -----  ---------------------------------------------------------------
claude       CAN_USE    5h     5h: 8% used (reset in 1h 39m), 7d: 22% used (reset in 5d 20h 39m) (all models), 7d: 15% used (reset in 5d 2h 39m) (sonnet only)
gemini/pro   CAN_USE    pro    4% used (reset in 14h 14m)
gemini/flash CAN_USE    flash  40% used (reset in 14h 18m)
copilot      LOW_QUOTA  -      72% used (reset in 9d 11h)
codex        CAN_USE    5h     5h: 65% used (reset in 3h), 7d: 21% used (reset in 6d)

JSON output example

ai-quota --json

{
  "claude": { "status": "ok", "reason": null, "error": null, "data": { ... }, "display": "5h: 8% used (...)" },
  "gemini": { "status": "ok", "reason": null, "error": null, "data": { ... }, "display": "pro: 4% used (...)" },
  "copilot": { "status": "ok", "reason": null, "error": null, "data": { ... }, "display": "72% used (...)" },
  "codex": { "status": "ok", "reason": null, "error": null, "data": { ... }, "display": "5h: 65% used (...)" }
}

Credential lookup

| Agent | Source | | ------- | -------------------------------------------------------------------------------------------------------------------------- | | Claude | ~/.claude/.credentials.json | | Gemini | ~/.gemini/oauth_creds.json | | Copilot | COPILOT_GITHUB_TOKEN, GH_TOKEN, GITHUB_TOKEN, Copilot CLI stored OAuth token, gh auth token, or legacy hosts.yml | | Codex | ~/.codex/auth.json |

Claude token behavior: ai-quota automatically refreshes Claude OAuth tokens using refreshToken when needed (including stale-token 401/403/429 retries). Copilot token behavior: if you already authenticated with copilot login, ai-quota will reuse that stored OAuth token automatically, including Windows Credential Manager entries created by Copilot CLI.

Exit code is 0 when the command successfully reports status output (table or JSON), even if some providers are unavailable/login-required/waiting for reset. Use --strict to make provider fetch errors fail with exit code 1. Invalid usage (unknown option/agent) and fatal command failures also exit with 1.

Advanced usage (SDK)

To fetch quota for all agents at once:

import { fetchAllRateLimits } from "@metyatech/ai-quota";

const all = await fetchAllRateLimits();
console.log("Overall status:", all.summary.status); // "healthy", "warning", or "critical"
console.log("Summary message:", all.summary.message);
console.log("Claude status:", all.claude.display);

To fetch only specific agents (more efficient):

import { fetchAllRateLimits } from "@metyatech/ai-quota";

// Only fetch Claude and Copilot
const results = await fetchAllRateLimits({
  agents: ["claude", "copilot"]
});

console.log(results.claude.display);
console.log(results.copilot.display);
console.log(results.gemini.display); // "skipped"

Supported agents

| Agent | Source | API type | | ------- | ------------------------------ | ----------------------- | | Claude | ~/.claude/.credentials.json | REST (Anthropic OAuth) | | Gemini | ~/.gemini/oauth_creds.json | REST (Google OAuth) | | Copilot | GitHub token (caller-provided) | REST (GitHub API) | | Codex | ~/.codex/auth.json | REST (ChatGPT internal) |

Requirements

• Node.js >= 18
• TypeScript (peer — types are included in the package)

Installation

npm install @metyatech/ai-quota

Usage

Claude

import { fetchClaudeRateLimits } from "@metyatech/ai-quota";

const usage = await fetchClaudeRateLimits();
if (usage) {
  console.log("5h utilization:", usage.five_hour?.utilization);
  console.log("7-day utilization:", usage.seven_day?.utilization);
}

Gemini

import { fetchGeminiRateLimits } from "@metyatech/ai-quota";

const usage = await fetchGeminiRateLimits();
if (usage) {
  const pro = usage["gemini-3-pro-preview"];
  console.log("Pro used %:", pro?.usage, "/ limit:", pro?.limit);
}

Set AGENT_RUNNER_GEMINI_OAUTH_CLIENT_ID and AGENT_RUNNER_GEMINI_OAUTH_CLIENT_SECRET when the Gemini CLI is not installed, or extract them from the Gemini CLI source.

Copilot

import { fetchCopilotRateLimits } from "@metyatech/ai-quota";

const usage = await fetchCopilotRateLimits({ token: process.env.GITHUB_TOKEN! });
if (usage) {
  console.log("Percent remaining:", usage.percentRemaining);
  console.log("Resets at:", usage.resetAt);
}

Options:

| Option | Type | Default | Description | | ---------------- | -------- | ------------------------ | ---------------------------- | | token | string | required | GitHub personal access token | | timeoutSeconds | number | 20 | Request timeout in seconds | | apiBaseUrl | string | https://api.github.com | Override GitHub API base URL | | apiVersion | string | 2025-05-01 | GitHub API version header |

Codex

import { fetchCodexRateLimits, rateLimitSnapshotToStatus } from "@metyatech/ai-quota";

const snapshot = await fetchCodexRateLimits({ codexHome: "~/.codex" });
const status = rateLimitSnapshotToStatus(snapshot);
const weekly = status?.windows.find((w) => w.key === "weekly");
console.log("Weekly % left:", weekly?.percentLeft);

Options for fetchCodexRateLimits:

| Option | Type | Default | Description | | ---------------- | ---------- | ---------- | ----------------------------------- | | codexHome | string | ~/.codex | Path to the Codex home directory | | timeoutSeconds | number | 20 | HTTP API request timeout in seconds | | timingSink | function | none | Callback for per-phase timing (ms) |

Dev commands

npm install       # install dependencies
npm run build     # compile TypeScript to dist/
npm test          # run tests with vitest
npm run lint      # ESLint + tsc typecheck
npm run format    # Prettier format
npm run verify    # lint + test + build (full CI suite)

Release and publish

ai-quota publishes to npm from GitHub Actions via npm trusted publishing (OIDC). The release workflow lives in .github/workflows/publish.yml, runs when a GitHub release is published, and uses Node.js 24 because npm trusted publishing requires a modern npm/Node runtime in CI.

Canonical publish path: create/publish a GitHub Release for a vX.Y.Z tag and let GitHub Actions publish to npm. Do not use long-lived NPM_TOKEN secrets for this package.

One-time npm setup (outside this repo):

1. In npm package settings for @metyatech/ai-quota, add a trusted publisher for metyatech/ai-quota with workflow .github/workflows/publish.yml and trigger environment matching published releases.
2. Keep that npm trusted publisher binding active; no npm automation token is required.

Release flow:

npm version patch --no-git-tag-version
# update CHANGELOG.md as needed
npm run verify
git add package.json package-lock.json CHANGELOG.md
git commit -m "Release vX.Y.Z"
git push origin main
git tag vX.Y.Z
git push origin vX.Y.Z

Then publish a GitHub release for the vX.Y.Z tag. GitHub Actions will run the publish workflow, publish the package to npm, and attach npm provenance automatically when trusted publishing is configured for the package.

Environment variables

| Variable | Used by | Purpose | | ----------------------------------------- | ------- | ----------------------------------------------- | | AGENT_RUNNER_GEMINI_OAUTH_CLIENT_ID | Gemini | Override OAuth client ID when Gemini CLI absent | | AGENT_RUNNER_GEMINI_OAUTH_CLIENT_SECRET | Gemini | Override OAuth client secret |

SemVer policy

Breaking changes (removed/renamed exports, changed function signatures) bump the major version. New exports and backward-compatible changes bump the minor version. Bug fixes bump the patch version.

Links

• CHANGELOG.md
• SECURITY.md
• CONTRIBUTING.md
• CODE_OF_CONDUCT.md
• LICENSE