CLI & MCP

The raw Agent API is plain HTTP, but the payment step asks your agent to sign an EIP-3009 authorization — which normally means installing x402 + viem, handling ESM/CJS, prefixing the key with 0x, and knowing the payload is x402Version 1 while the challenge is version 2. @xintel/agent bakes all of that in. It’s one package that runs as both a CLI and an MCP server, so an agent can buy a report in a single command with no crypto toolchain to bootstrap.

Buys shared report bodies only — never profile, posts, or edges. Same product and pricing as the HTTP API ($0.01 USDC per report on Base).

CLI (npx)

Nothing to install — npx fetches and caches it:

# free browsing (no wallet)
npx @xintel/agent list
npx @xintel/agent menu --username AskVenice

# paid ($0.01 USDC per report on Base)
export XINTEL_AGENT_KEY=0x<base-wallet-private-key-funded-with-usdc>
npx @xintel/agent buy --username AskVenice --select all
npx @xintel/agent buy --username AskVenice --select range --ids <id1>,<id2>

Add --json to any command for the raw response. The buy command runs the full 402 → sign → retry loop and prints the delivered reports.

MCP server

Add Xintel to any MCP-aware host (Cursor, Claude Desktop) — the same package, started with the mcp subcommand:

{
  "mcpServers": {
    "xintel": {
      "command": "npx",
      "args": ["-y", "@xintel/agent", "mcp"],
      "env": { "XINTEL_AGENT_KEY": "0x<base-wallet-private-key>" }
    }
  }
}

Tools exposed:

Tool	Paid?	Purpose
`xintel_list`	free	list usernames with shared reports
`xintel_menu`	free	list a username’s report menu (ids, dates, model)
`xintel_purchase`	`$0.01`/report	buy report bodies — handles the 402/sign/retry loop internally

Environment

Variable	Required for	Description
`XINTEL_AGENT_KEY`	`buy` / `xintel_purchase`	Base wallet private key funded with USDC. Read from env only — never a CLI flag. The client runs locally; your key never touches Xintel’s servers.
`XINTEL_BASE_URL`	optional	Override the API base (default `https://xintel.aispace.bot`; use `http://localhost:5173` for local dev).

When to use what

MCP — best for hosted agents (Cursor, Claude Desktop). One-time mcp.json entry, then the agent calls tools directly.
CLI (npx) — best for scripts, cron, or agent frameworks (LangChain, CrewAI) that shell out. Zero setup per run.
Raw HTTP — see the Quickstart if you’d rather implement the flow yourself or can’t run Node.

All three hit the same endpoints and the same $0.01/report pricing.

Overview

Using the app

Agent API

Access models

Reference

CLI (npx)

MCP server

Environment

When to use what

​CLI (npx)

​MCP server

​Environment

​When to use what

CLI (npx)

MCP server

Environment

When to use what