Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.costhawk.ai/llms.txt

Use this file to discover all available pages before exploring further.

CostHawk

​
What is CostHawk?

CostHawk is an AI cost platform for teams using OpenAI, Anthropic, Google, and coding assistants. Use it in progressive depth:
  • Admin API sync for low-friction org reporting
  • OTel for real-time telemetry
  • Proxy + wrapped keys for hard-stop enforcement
  • MCP for assistant-native workflows (fully supported)

​
Key Features

Usage Tracking

Monitor API calls, tokens, and costs across all your AI providers in real-time.

Savings Calculator

See if your flat-rate subscriptions (Claude Max, OpenAI Pro) are worth it.

Alerts & Anomalies

Get notified about cost spikes, budget warnings, and unusual activity.

Admin API Sync

Connect provider admin keys to ingest org-level usage and costs quickly.

API Proxy Controls

Add wrapped-key routing for runtime controls and hard spending limits.

MCP Server

Query your CostHawk data directly from Claude Code, Claude Desktop, or Codex workflows.

Trust & Privacy

Local-first parsing, no prompt content, preview before upload, opt-in auto-sync.

​
Quick Start

Get started in under 5 minutes:
1

Choose your path

Start with Admin API Sync for quick visibility, then layer on Proxy/OTel as needed.
2

Enable MCP for assistant workflows

Install CostHawk MCP so you can query usage/savings directly from your coding assistant.
3

Install MCP with Auto-Sync (optional but recommended)

All platforms (recommended):
npm exec --yes costhawk@latest -- --login
This handles browser approval + setup automatically.
4

Start tracking

Restart your MCP host, then ask: “What’s my AI API usage this month?”

​
Why CostHawk?

  • Admin API sync - Pull provider usage/cost data with minimal setup
  • API proxy - Add per-request enforcement and hard-stop controls
  • Multi-provider support - Track OpenAI, Anthropic, Google, and more in one place
  • Flat-rate savings analysis - Know if Claude Max or OpenAI Pro is worth it for you
  • Spending limits - Set daily/monthly budgets with hard stops to prevent runaway costs
  • MCP integration - Query your costs directly from assistants (fully supported)
  • Real-time alerts - Never be surprised by an API bill again
  • Privacy-first design - Local parsing, no prompt content, preview before upload
  • Developer-first - Built by developers, for developers

​
Trust & Privacy

We believe transparency builds trust. CostHawk is designed so you can verify everything:
Run npm exec --yes costhawk@latest -- --what-we-read to see exactly which files the MCP server accesses. We only read usage metadata from ~/.claude/projects and Codex session logs (~/.codex/sessions, ~/.config/codex/sessions, or ~/.local/share/codex/sessions).
Only usage metadata: token counts, model names, timestamps, and hashed project IDs. Never your code, prompts, or file paths. Use dryRun: true to preview any payload.
Auto-sync is off by default. Enable with COSTHAWK_AUTO_SYNC=true. You can also disable Codex sync specifically with COSTHAWK_CODEX_AUTO_SYNC=false.