MCP in Cursor & IDEs - Abstraxn Docs

Use this guide when you want IDE-native access to Agent Kit tools (balance, transfer intent, x402 discovery) without building a full app first. For a terminal REPL with chat and MCP tools, see Abstraxn CLI. For ChatGPT, Claude Web, or any cloud LLM connector, see Connect external agents. For LLM integration inside your own backend, see Full-stack app integration.

MCP-only quick path: Dashboard setup → 5-minute smoke test → paste config below → restart Cursor. Signing (transfer, paid_fetch payment) still needs a backend with accessKey — see Credentials.

Prerequisites

Complete Dashboard setup.
Copy API Key and MCP URL from Agentic Stack → Overview.
Optional: run Smoke test with the same key before configuring Cursor.

Cursor configuration

Paste into your Cursor MCP config (mcp.json or Cursor settings), replacing the API key:

{
  "mcpServers": {
    "web3-agent-kit": {
      "url": "https://agent-kit.abstraxn.com/mcp",
      "transport": "http",
      "headers": {
        "Authorization": "<application-api-key-from-dashboard>"
      }
    }
  }
}

Dev environments may use:

https://dev-agent-kit.abstraxn.com/mcp

Restart Cursor after saving. The MCP panel lists whatever tools/list returns — names change as Agent Kit adds tools.

Claude Code

Use a per-agent API key (recommended) so you do not need agent_id in every tool call:

claude mcp add -s user web3-agent-kit https://agent-kit.abstraxn.com/mcp \
  --header "Authorization: YOUR_PER_AGENT_API_KEY"

Verify with claude --mcp-debug. See Connect external agents for ChatGPT and Claude Web (OAuth). If the Claude connector icon looks wrong, see Connector icon.

Application key vs per-agent key

Key	When to use in Cursor
Application API key (Overview)	Default; pass `agent_id` in tool args if you have multiple agents
Per-agent API key (`createAgent` → `agent.apiKey`)	Set as `Authorization`; omit `agent_id`

Create agents via SDK quickstart or REST POST /agents before calling wallet tools.

Calling tools manually

curl -s -X POST "$MCP_URL" \
  -H "Authorization: $APP_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "<from-tools-list>",
      "arguments": { "agent_id": "YOUR_AGENT_UUID" }
    }
  }'

x402 and signing in IDEs

Catalog discovery tools (from tools/list) — typically no wallet signing.
Paid / x402 tools — require paymentPayload after a 402 response; signing needs the agent access key on a backend (x402 payments).
Transfer intents — return unsigned tx; signing requires access key (not doable from Cursor alone unless you run a local signer script).

For autonomous signing, use @abstraxn/agent-kit in a small Node script or the full-stack sample.

Autonomous transactions from IDEs

When an agent has autonomous transactions enabled, external MCP callers (Cursor, CLI) receive transaction_pending instead of only unsigned_transaction_ready. Your integrator backend must claim, sign, and broadcast queued transactions (webhook or poll). In-app chat integrations should send X-Agent-Kit-Execution-Context: managed and sign immediately on the server.

Tool reference

See MCP tools reference for every tool, arguments, and response shapes.

​Prerequisites

​Cursor configuration

​Claude Code

​Application key vs per-agent key

​Calling tools manually

​x402 and signing in IDEs

​Autonomous transactions from IDEs

​Tool reference

Prerequisites

Cursor configuration

Claude Code

Application key vs per-agent key

Calling tools manually

x402 and signing in IDEs

Autonomous transactions from IDEs

Tool reference