Using Steel.dev with prxy.monster

Steel runs cloud browsers and a scrape API. PRXY routes and compresses LLM calls. They are complementary:

Quick stack

# LLM → PRXY
export ANTHROPIC_BASE_URL=https://api.prxy.monster
export ANTHROPIC_API_KEY=prxy_live_xxx
 
# Steel → gateway env (steel-browser module)
export STEEL_API_KEY=steel_xxx
 
# Pipeline with scrape + compression
export PRXY_PIPE=steel-browser,mcp-optimizer,structured-crusher,code-crusher,ipc,patterns

steel-browser module

PRXY ships steel-browser — injects steel_scrape and fulfills it server-side. Your agent only talks to PRXY; no Steel SDK in the client unless you want full browser sessions.

curl https://api.prxy.monster/v1/messages \
  -H "Authorization: Bearer $PRXY_API_KEY" \
  -H "x-prxy-pipe: steel-browser,structured-crusher,ipc,patterns" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "tools": [],
    "messages": [{"role":"user","content":"Summarize https://example.com in 3 bullets"}]
  }'

The module injects steel_scrape; the model calls it; PRXY hits Steel and auto-loops on non-streaming requests.

Mastra

Steel documents a Mastra cookbook. Point Mastra’s model at PRXY the same way as our Mastra integration:

export ANTHROPIC_BASE_URL=https://api.prxy.monster
export ANTHROPIC_API_KEY=prxy_live_xxx
export PRXY_PIPE=steel-browser,structured-crusher,ipc,patterns

Use PRXY’s steel-browser for scrape-in-pipeline, or Steel’s native browser tools for Playwright sessions — many teams use both (Steel sessions in the agent, PRXY on every LLM call).

LangGraph / Claude Agent SDK / Browser Use

Steel integrates with LangGraph, Claude Agent SDK, and Browser Use. Pattern is identical: Steel for browser, PRXY base URL for the LLM.

Self-hosted

Steel self-hosts via Docker. PRXY local mode airgaps LLM routing. Run both on your infra for full data control.

Example repo

Runnable reference: prxy-monster-examples/examples/steel-research-agent