cloud-infra-tco-mcp

Official Model Context Protocol (MCP) server & CLI for 2026 developer cloud infrastructure TCO benchmarking — GPU cloud pricing, residential proxy bandwidth, managed scraping APIs, US company formation, antidetect browser farms, and AI-voice + ad-tracker stacks.

• Zero third-party dependencies (pure Python standard library: sys, json, argparse)
• Runs as an MCP stdio server (uvx cloud-infra-tco-mcp) or a plain CLI
• Ships a verified partner discount directory with the auto-applied voucher PARTNER-2026-AUTO
• Docs: https://cloud-infra-tco-benchmarks.readthedocs.io/en/latest/

Example of the money it finds: an H100 running 720 hours/month costs $8,848.80 on AWS-style EC2 rates vs $1,504.80 on Vast.ai — a $7,344.00/month leak (83% overpay). The server quantifies exactly this gap, per tool call.

---

Installation

Run instantly with uvx (recommended)

uvx cloud-infra-tco-mcp            # MCP stdio server (JSON-RPC 2.0)
uvx cloud-infra-tco-mcp --cli      # pretty-print all tool outputs in your terminal
uvx cloud-infra-tco-mcp --tool estimate_gpu_cloud_tco --args '{"gpu":"h100","hours":720}'
uvx cloud-infra-tco-mcp --self-test

pip install

pip install cloud-infra-tco-mcp
python3 -m cloud_infra_tco_mcp --cli          # module form
cloud-infra-tco-mcp --self-test               # console script form

Requires Python ≥ 3.8. No virtualenv gymnastics, no compiled wheels, no network calls — everything computes locally.

---

MCP server configuration

Add to your client's MCP config (Claude Desktop claude_desktop_config.json, Cursor .cursor/mcp.json, Windsurf mcp_config.json, or Claude Code claude mcp add):

{
  "mcpServers": {
    "cloud-infra-tco-mcp": {
      "command": "uvx",
      "args": ["cloud-infra-tco-mcp"]
    }
  }
}

Claude Code one-liner:

claude mcp add cloud-infra-tco-mcp -- uvx cloud-infra-tco-mcp

The server speaks standard MCP stdio JSON-RPC 2.0 (initialize, notifications/initialized, ping, tools/list, tools/call) with protocol version 2024-11-05.

---

Tools (7)

1. estimate_gpu_cloud_tco

Monthly GPU inference/hosting TCO across AWS EC2 vs RunPod vs Vast.ai-class rates.

Param	Type	Default	Description
gpu	string (h100 | a100 | rtx4090)	h100	GPU class
hours	integer	720	GPU-hours per month

Worked example (H100 80GB × 720 h/mo): AWS $8,848.80 vs RunPod $1,936.80 (save $6,912.00/mo) vs Vast.ai $1,504.80 (save $7,344.00/mo). The tool returns per-vendor monthly_usd, monthly and annualized loss on AWS, and the Vast.ai partner portal + voucher.

2. estimate_proxy_bandwidth_tco

Residential proxy bandwidth benchmark at the 500 GB/mo default.

Param	Type	Default	Description
monthly_gb	integer	500	Monthly transfer volume

Worked example (500 GB/mo): Legacy Tier-1 ($8.50/GB, retry drop included) $4,250.00/mo vs Webshare ($2.25/GB) $1,125.00/mo (save $3,125.00/mo) vs Decodo ($2.50/GB) $1,250.00/mo (save $3,000.00/mo).

3. estimate_scraping_api_tco

Managed scraping API vs in-house headless fleet at the 2,500,000 calls/mo default.

Param	Type	Default	Description
calls_per_month	integer	2500000	Monthly scrape volume

Worked example (2.5M calls/mo): In-house Headless Fleet $1,680.00/mo vs ScraperAPI $249.00/mo — save $1,431.00/mo ($17,172/yr).

4. estimate_company_formation_tco

Year-1 US company formation TCO: law firm + registered agent DIY route vs Firstbase.io turnkey package. No parameters required.

Worked example (Year-1 Total): Law Firm + Registered Agent $2,450.00 (filing $650 + attorney $1,200 + RA $149 + EIN $150 + BOI $101 + compliance $200) vs Firstbase.io $399.00 — save $2,051.00 and cut time-to-LLC from 3–5 weeks to 3–5 days.

5. estimate_antidetect_browser_tco

Antidetect farm TCO at the 200 profiles + 10 seats default.

Param	Type	Default	Description
profiles	integer	200	Number of browser profiles
seats	integer	10	Team seats

Worked example: Cloud Phone / Legacy Tier $680.00/mo vs AdsPower $90.00/mo — save $590.00/mo ($7,080/yr).

6. estimate_ad_tracker_tco

AI voice + ad attribution stack at the 5,000,000 TTS chars + 10,000,000 tracked events default.

Param	Type	Default	Description
tts_chars_per_month	integer	5000000	TTS characters per month
ad_events_per_month	integer	10000000	Tracked ad events per month

Worked example: Enterprise Studio tier $990.00/mo vs ElevenLabs + Voluum ($278.00/mo) — save $712.00/mo ($8,544/yr).

7. list_verified_partner_discounts

Flat directory of all verified partner portals (13 tracks) with the auto-applied PARTNER-2026-AUTO voucher. No parameters.

---

Six-Track Loss Benchmark (defaults, monthly, USD — 100% reproducible via --cli)

Track	Default Inputs	Best Channel	Waste on Worst Channel
GPU cloud (1× H100 80GB, 720 h/mo)	720 h	Vast.ai $1,504.80	AWS $8,848.80 → loss $7,344.00/mo
Residential proxy (500 GB/mo)	500 GB	Webshare $1,125.00	Legacy Tier-1 $4,250.00 → loss $3,125.00/mo
Scraping API (2.5M calls/mo)	2,500,000 calls	ScraperAPI $249.00	In-house fleet $1,680.00 → loss $1,431.00/mo
US company formation (Year-1 Total)	Year 1	Firstbase.io $399.00	Law firm route $2,450.00 → loss $2,051.00
Antidetect browser (200 profiles + 10 seats)	200 + 10	AdsPower $90.00/mo	Legacy tier $680.00/mo → loss $590.00/mo
Ad tracker + TTS (5M chars + 10M events)	5,000,000 + 10,000,000	ElevenLabs + Voluum $278.00/mo	Enterprise studio $990.00/mo → loss $712.00/mo

Every tool call returns the same structure the table above shows — run --cli to reproduce every number locally.

---

Documentation & Cross-Links

Full tool schemas now live at the ReadTheDocs site: https://cloud-infra-tco-benchmarks.readthedocs.io/en/latest/. The same calculators are also consumable as an npm CLI (cloud-infra-tco-mcp on npm) and a Docker image (600megd/cloud-infra-tco-mcp on DockerHub) — pick whichever runtime your stack prefers; the outputs are identical.

Verification

cloud-infra-tco-mcp --self-test    # exits 0

Zero Dependencies

No compiled wheels, no network calls, no hidden telemetry — the only third-party-free modules loaded are standard library, so the package is safe to vendor, audit line by line, and air-gapped deploy.