ArgusArgus
Start free

Let your coding agent watch your API spend.

Argus gives Claude Code, Cursor and Codex a live read on every metered API you build with. Mid-session, your agent can check what a run cost, catch a loop burning credits, and notice a provider running dry.

And you get one place for every balance, spend line and runway, with an email before one hits zero.

Free up to $1,000 a month of tracked spend. No card.

Every meter, in one place.

Consumption pricing put a meter on everything you build with — credits, tokens, calls, rate limits. The providers all bill flawlessly; none of them tell you where you stand. So you check ten dashboards, or you check none and find out when production stops.

We found out the second way. The week Argus started, three of our own meters were already dry.

Paid metersone afternoon in June
OpenRouterbalance −$0.11runway: 0 days
Firecrawl0 credits leftevery crawl fails
Tavily1,003 / 1,000over plan

Real numbers. Argus read all three in one poll.

A URL, a key, or a line of code.

MCP for the agent doing the spending, polling for authoritative totals, a fetch wrapper for per-request detail.

"url": "https://api.argus.town/mcp"
→ runway, spend, limits, per-request costs

Point your agent

Over MCP, your coding agent can check what a run cost, find the loop burning credits, and steer around a provider that's about to cut it off.

openrouter · sk-or-••••••••••••
→ balance · usage · plan, hourly

Paste a key

Connect a provider and Argus polls its billing endpoints on your behalf: authoritative totals, remaining credits, plan status. No code.

globalThis.fetch =
  argus.instrument(fetch);

Wrap fetch

One line at your app's edge. Every call gets itemized as it happens: cost, latency, and the rate-limit headers billing pages never show.

Or hand it to your agent — paste this into Claude Code, Cursor, or any harness:

What you see once a key or the SDK is in.

Every request, itemized.

Cost comes from the provider when it reports one, and from a synced catalog of 1,000+ model rates when it doesn't. Each row says which, and the total reconciles against the provider's own bill.

Streaming from @argus/sdk
Recent requests with per-request cost and its source
TimeRequestCost
14:02:37claude-haiku-4.5AnthropicDerived$0.010312
14:02:34/searchExaReported$0.005000
14:02:31gpt-4.1-miniOpenAIDerived$0.002371
14:02:28anthropic/claude-sonnet-4.5OpenRouterReported$0.009402
14:02:25/searchTavilyDerived$0.008000
14:02:22deepseek/deepseek-chatOpenRouterReported$0.001790
14:02:19/v1/scrapeFirecrawlDerived$0.005330
14:02:16claude-haiku-4.5AnthropicDerived$0.010312
14:02:13/searchExaReported$0.005000
14:02:10gpt-4.1-miniOpenAIDerived$0.002371
Exact — itemized costs reconcile with the providerOpenRouter this month: itemized $1,284.07, billed $1,289.21Drift 0.4%

A countdown on every balance.

Burn rate from recent spend, projected to zero.

OpenRouterBalance
$211.60
30 days agoTodayZero
Burn rate
$11.62/day
Days to zero
18 days

Headroom before the 429.

Read from the rate-limit headers on your real traffic.

OpenAIgpt-4.1
Requests per minute0 / 5,000
Tokens per minute1.71M / 2M
429s in the last hour3
x-ratelimit-remaining-requests: 390

Your agent can check the meter.

Point any MCP client at https://api.argus.town/mcp. Read tools return runway, spend, rate limits and per-request costs.

→ tools/call get_rate_limit_status← [    { "provider": "openai",      "headroomPct": 8, "errorRatePct": 2.1 },    { "provider": "anthropic",      "headroomPct": 64, "errorRatePct": 0 }  ]# openai is near its limit; batch goes to anthropic
  • get_overview
  • get_rate_limit_status
  • get_request_costs
  • get_cost_accuracy
  • get_advisories
  • get_provider
  • list_connections
  • list_available_providers

An email while there's time to act.

Set a threshold once. Argus checks it every hour.

When runway falls below 7 days on DeepSeekEnabled
Argus alert: deepseek09:00

DeepSeek (deepseek): Credits run out in ~6 days (threshold 7).

  • OpenAI (openai): Rate-limit headroom 8% at or below 10%.2d ago
  • Exa (exa): Spend up 64% week-over-week (threshold 50%).5d ago

Also on balance, spend spikes, projected overage, rate-limit headroom, 429 rate and cost drift.

The plan that fits your usage.

Argus compares this cycle against the provider's tiers and tells you when to move.

FirecrawlHobby plan
Credits this cycle0 / 3,000

Resets in 9 days. At this pace you need about 4,100.

Upgrade to Standard+$67/mo

LLMs, and the long tail.

28 providers connect with a key. Anything else reports through the SDK or one signed POST.

  • OpenAI
  • Anthropic
  • OpenRouter
  • DeepSeek
  • Gemini
  • Perplexity
  • Exa
  • Firecrawl
  • Tavily
  • Bright Data
  • Apify
  • Apollo
  • Hunter
  • Prospeo
  • LeadMagic
  • People Data Labs
  • Crunchbase
  • Crustdata
  • Parallel
  • Postmark
  • SendGrid
  • MillionVerifier
  • SecurityTrails
  • WhoisXML

Provider keys are AES-256 encrypted at rest, never logged, and never returned by the API.

Stop guessing what you're spending.

Connect a provider key and point your agent at the MCP server. Free up to $1,000 a month of tracked spend.