MCP-native API Broker

Your agents overpay
for every API call.
We fix that.

Hebline routes every API call — including LLM calls — to the best service at the right price. Free when it's enough. Paid when it matters. It knows the difference.

The difference

Every other router earns
when you pay more.

OpenRouter, LiteLLM, Portkey — they take a margin on paid calls. Routing you to free alternatives kills their revenue. Our model doesn't depend on your spend — so we're the only router built to actually save you money.

No margin on your API calls. Ever.

The problem

Your agents are bleeding money.

LLM calls are the biggest leak

Sonnet for a yes/no classification. GPT-4o for a regex match. Your agent doesn't know it could use Gemini Flash for free — because no one told it.

Cascading API costs

One agent task triggers 5–10 paid API calls across different providers. Token-based, request-based, tiered — no unified cost unit.

False savings are as dangerous as overspending

Using a free geocoder for a payment-critical address check isn't smart. It's a risk. The question isn't free or paid — it's what does this call actually require?

How it works

Three steps. Zero friction.

Step 1

Agent describes need

> "Classify this support
ticket as urgent or not"

The agent says what it needs. No model selection, no API docs.

Step 2

Hebline compares & routes

Scoring models...

$ claude-sonnet $0.003/call

F gemini-flash free — score 0.91

→ gemini-flash (best match)

Cost, quality, latency — scored in one pass. Free when the task allows it.

Step 3

Agent gets result

✓ Classification: not urgent

Model: Gemini Flash

Latency: 180ms

Cost: $0.00

Free when the task allows it. Paid when precision matters. The agent notices no difference.

Install

One command. Start saving.

$npm install -g @hebline.ai/mcp-server
claude_desktop_config.json
{
  "mcpServers": {
    "hebline": {
      "command": "npx",
      "args": ["-y", "-p", "@hebline.ai/mcp-server", "hebline-mcp"]
    }
  }
}
Agent call

> "Classify this support ticket as urgent or not"

// Hebline evaluates: Claude Sonnet ($0.003), Gemini Flash (free)...

// Routes to: Gemini Flash (free, 91% quality match)

✓ Classification: not urgent

Cost: $0.00 | Latency: 180ms | Model: Gemini Flash

Coming soon

ChatGPTGoogle GeminiGrokCody (Sourcegraph)Replit

The promise

Precise Routing.
Free when possible. Paid when necessary.

Most calls don't need the best model. Some do. Hebline learns precisely when it matters — and keeps learning as the market changes. You ship once and always pay the right price.

Every category. Always the right tier.

LLMs

Live
PaidOpenAI GPT-4o
FreeGroq (Llama 3.3) / Gemini Flash

Geocoding

Live
PaidGoogle Maps
FreeNominatim

Translation

Live
PaidDeepL
FreeMyMemory

Web Scraping

Live
PaidFirecrawl
FreeFetch Scraper

Currency

Live
PaidFixer.io
FreeExchangeRate-API

OCR

Live
PaidGoogle Vision
FreeOCR.space

Weather

Live
PaidOpenWeatherMap
FreeOpen-Meteo

Web Search

Live
PaidBrave Search
FreeDuckDuckGo

News

Live
PaidNewsAPI.org
FreeHackerNews
and many more ...

Architecture

How Hebline knows when free is enough.

Layer 1

Discovery

"What can handle this task?"

Your agent says what it needs — Hebline finds matching services across all categories. API docs, pricing and capabilities are indexed for instant matching.

Layer 2

Scoring

"Not just the cheapest. The most appropriate."

Cost, quality, latency, reliability — scored in one pass. When the router is confident free is enough, it routes free. When it's not confident, it pays.

Layer 3

Learning

"After 1,000 calls, it just knows."

Every call is a data point. Success rates per model, per task type, per complexity level. The router gets more confident — and your savings grow.

What we've learned so far

Live data from the routing layer.

73%

of calls routed to free alternatives

$0.00

average cost for non-critical tasks

9

categories live

<200ms

average routing decision

MCP gave agents a language.
Hebline gives them judgment.

Your agents will never overpay. And never cut corners where it counts.