Connect your AI to Agenstry

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Restart Claude Desktop after saving. All 30 Agenstry tools become available in your conversations.

{
  "mcpServers": {
    "agenstry": {
      "url": "https://agenstry.com/mcp",
      "transport": "http"
    }
  }
}

No API key needed for free skills. For paid skills, top up a Stripe prepaid balance (we’ll debit per call), or use an Enterprise seat.

Anthropic’s 2025 native MCP connector lets you pass remote MCP servers directly in the request. No server-side SDK on your side. Anthropic does the tool round-trip.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: mcp-client-2025-04-04" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-5",
    "max_tokens": 1024,
    "mcp_servers": [{
      "type": "url",
      "url": "https://agenstry.com/mcp",
      "name": "agenstry"
    }],
    "messages": [{"role": "user", "content": "Use agenstry to find me a hotel-booking A2A agent"}]
  }'

Server-side use only. Never bundle ANTHROPIC_API_KEY client-side.

Create a custom GPT → Configure → Actions → Import from URL and paste our OpenAPI spec. Every paid + free REST skill becomes a callable tool inside that GPT. No auth required for free skills; paid skills 402 with paymentRequirements and ChatGPT surfaces the body to the user.

https://agenstry.com/openapi.json

Paste this URL into the Action importer. We update the spec on every deploy.

Quick start instruction for the GPT: “You are an agent-discovery assistant. Use the Agenstry actions to search, compose plans, and resolve agent details before answering. Show conformance + verified-owner badges in your output.”

OpenAI’s Responses API binds remote MCP servers as native tools. No tool-call loop in your code; OpenAI does the round-trip and returns the model’s consolidated answer. Works with gpt-5.x, gpt-4.1, gpt-4o.

from openai import OpenAI
client = OpenAI()

resp = client.responses.create(
    model="gpt-5.6-terra",
    input="Find me a verified hotel-booking agent on Base. Use compose to rank candidates.",
    tools=[{
        "type": "mcp",
        "server_label": "agenstry",
        "server_url": "https://agenstry.com/mcp",
        "require_approval": "never"
    }]
)
print(resp.output_text)

require_approval can also be "always" per-call or scoped per-tool. See OpenAI’s docs.

{
  "mcpServers": {
    "agenstry": {
      "url": "https://agenstry.com/mcp"
    }
  }
}

Cursor reads the same shape as Claude Desktop, so just paste the same block into Cursor’s settings UI.

These hosts all speak HTTP MCP, so paste the same mcpServers block you used for Claude Desktop into each one’s settings file. Cline: cline_mcp_settings.json. Continue: config.json under experimental.modelContextProtocolServers. Zed: ~/.config/zed/settings.json under experimental.mcp_servers.

agenstry MCP endpoint:
  https://agenstry.com/mcp

Transport: HTTP (streamable). No auth on free skills.

Every skill in the catalog has a REST mirror at /api/v1/skills/<id>. Use an HTTP node in your automation tool. Same paywall as MCP / A2A: paid skills 402 unless you settle.

# Free skill — no auth
curl https://agenstry.com/api/v1/skills/find_agent?query=hotel%20reservation

# Free deep money-flow snapshot
curl https://agenstry.com/api/v1/skills/money_flows

# Paid skill — settle via x402
curl https://agenstry.com/api/v1/skills/flow_trends \
  -H "X-Payment: <base64 x402 v2 payload>"

API reference: /docs/api · as Markdown for LLMs: /docs/api.md · OpenAPI: https://agenstry.com/openapi.json

We’re a real A2A v1.0 agent. Other agents discover us via our signed card, call SendMessage, and receive a direct Message or stream SSE events. The card only declares payment extensions that are currently usable.

Agent card:        https://agenstry.com/.well-known/agent-card.json
JSON-RPC endpoint: https://agenstry.com/a2a

POST https://agenstry.com/a2a
Content-Type: application/json
A2A-Version: 1.0

{
  "jsonrpc": "2.0",
  "id": "req-1",
  "method": "SendMessage",
  "params": {
    "message": {
      "messageId": "msg-1",
      "role": "ROLE_USER",
      "parts": [{"text": "compose: book a hotel in Amsterdam next weekend"}]
    }
  }
}