AI Agent Tool Discovery: How MCP and Lightning Enable Pay-Per-Use AI

AI agents need two things to use tools autonomously: a way to discover what's available, and a way to pay for it. MCP handles discovery. Lightning handles payment. Together, they let agents find and use 25+ AI tools without API keys, subscriptions, or human intervention.

The Problem with Tool Access Today

How Tool Discovery Works

MCP defines a standard way for AI agents to interact with external tools. Instead of custom integrations per service, agents connect to an MCP server that exposes tools through a common protocol. Discovery happens at three levels:

What changes when agents can discover and pay for tools on their own

How Agents Handle Long-Running Tasks

Some AI operations take time — video generation, 3D model creation, voice cloning. Rather than blocking the agent, these tools follow an async lifecycle that lets agents fire off work and check back later.

Every async response includes the jobType so agents don't need to track it themselves. The response also includes recommended polling intervals and maximum wait times, so agents can decide when to give up.

Technical details for developers integrating MCP tools

Discover Available Tools

# Static discovery — what tools exist
curl https://sats4ai.com/.well-known/mcp | jq '.tools[] | .name'

# Live catalog — tools with current pricing
curl https://sats4ai.com/api/mcp/discovery | jq '.tools[] | {name, pricing}'

# MCP handshake — full tool schemas
curl -X POST https://sats4ai.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Pay and Execute

# Step 1: Create payment
curl -X POST https://sats4ai.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"create_payment",
                 "arguments":{"toolName":"image"}}}'
# Returns: { paymentId, invoice, amountSats }

# Step 2: Pay the Lightning invoice (milliseconds)

# Step 3: Execute the tool
curl -X POST https://sats4ai.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"image",
                 "arguments":{"paymentId":"<id>",
                              "prompt":"A neon-lit Tokyo alley"}}}'
# Returns: base64 image

Async Tools (Video, 3D, Voice Clone)

# Submit async job (video example)
# Returns: { requestId, jobType: "video", status: "IN_PROGRESS" }

# Poll status
curl -X POST https://sats4ai.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
       "params":{"name":"check_job_status",
                 "arguments":{"requestId":"<id>",
                              "jobType":"video"}}}'
# Returns: { status: "IN_PROGRESS", next: "Poll again in 5 seconds." }
# or:      { status: "COMPLETED", next: "Call get_job_result..." }

# Retrieve result
curl -X POST https://sats4ai.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call",
       "params":{"name":"get_job_result",
                 "arguments":{"requestId":"<id>",
                              "jobType":"video"}}}'
# Returns: { videoUrl: "..." }

MCP & OpenClaw

Orchestrators, agents, and the tool layer

Most production agent systems aren't a single agent talking to a single tool. They're orchestrators — higher-level systems that decompose a goal into subtasks and route them to specialized agents or tools. The orchestrator handles task decomposition, agent routing, state management, and error recovery. The agents handle execution.

Sats4AI is the tool layer both single agents and orchestrated multi-agent systems call into. A single Claude agent calls us directly for an atomic task. A LangGraph pipeline routes its comms sub-agent to our SMS endpoint and its media sub-agent to our image generation endpoint. Either way, each tool call is a single MCP request with a Lightning payment — no per-agent configuration, no shared credentials.

25+ Tools Available via MCP