QuickstartMCPIntegration

5-minute quickstart: AstraNL MCP in Claude Desktop, Cursor, CrewAI, LangChain

29 April 2026 · 4 min read

AstraNL is a single MCP endpoint that aggregates 15 AI capabilities (translate, write, code, summarise, classify, extract, transform, plan, compare, brainstorm, proofread, qa, explain, analyze, document_qa) and routes them across Anthropic, OpenAI, Gemini, and Grok with automatic failover. Pricing is real provider cost passed through plus a 15% infrastructure markup; you fund a prepaid wallet once and call as much as you want.

Here is how to plug in.

Step 0 — Register an agent and fund a wallet

# Register — returns X-Agent-Key (save it, this is your auth)
curl -s https://astranl.com/api/agents/register \
  -H 'Content-Type: application/json' \
  -d '{"name":"my-agent"}' | jq

# Topup wallet — minimum €5, returns Stripe Checkout URL
curl -s https://astranl.com/api/agents/wallet/topup \
  -H 'X-Agent-Key: ask_...' \
  -H 'Content-Type: application/json' \
  -d '{"amount_cents":500}' | jq

Pay the Checkout URL once. Wallet credits arrive via webhook in seconds.

Step 1 — Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your OS:

{
  "mcpServers": {
    "astranl": {
      "url": "https://astranl.com/mcp/sse",
      "transport": "sse"
    }
  }
}

Restart Claude Desktop. AstraNL tools now appear in the tool picker.

Step 2 — Cursor

Settings → MCP → Add new server, then paste:

{
  "name": "astranl",
  "url": "https://astranl.com/mcp/sse",
  "type": "sse"
}

Step 3 — CrewAI

from crewai import Agent
from crewai_tools import MCPTool

agent = Agent(
    role="researcher",
    goal="summarize content across languages",
    tools=[MCPTool(
        url="https://astranl.com/mcp/sse",
        headers={"X-Agent-Key": "ask_..."}
    )]
)

Step 4 — LangChain

from langchain_mcp import MCPClient

client = MCPClient(
    server_url="https://astranl.com/mcp/sse",
    headers={"X-Agent-Key": "ask_..."}
)
tools = client.list_tools()
result = client.call_tool("translate",
    {"text": "Hello", "target_language": "nl"})

Step 5 — Verify it works (HTTP, no SDK)

curl -s https://astranl.com/capabilities/execute \
  -H 'X-Agent-Key: ask_...' \
  -H 'Content-Type: application/json' \
  -d '{"capability":"translate",
        "input":{"text":"Coordination protocol for AI agents",
                 "target_language":"nl"}}'

What you get

Multi-provider failover (Anthropic primary, OpenAI / Gemini / Grok fallback)
Micro-EUR per-call billing visible in GET /api/agents/wallet/balance
Public tool list at /mcp/tools.json (HTTP, no SSE handshake required)
Working examples (Python, Node, LangChain, CrewAI) at github.com/tolegm/astranl-mcp-examples

What you should not expect

Free inference — we pass through real provider cost. €5 wallet ~= 4348 translate calls or ~435 code-generation calls; we do not subsidise.
Local execution — AstraNL is a remote SSE endpoint, not a stdio server. There is no npx form.

If you hit anything that does not match this guide, ping hello@astranl.com or open an issue at github.com/tolegm/astranl-mcp. We read everything.

—

More AstraNL insights