Give Claude Code Auto Mode 318 Paid APIs

Claude Code auto mode goes default on August 14. Your agent will autonomously execute tool calls — no permission prompts, no "Allow? (y/n)" interruptions. But which tools will it have access to?

The answer that ships with Claude Code: filesystem, bash, grep, web search. Useful, but limited. What if your agent could also query live gas prices across 8 chains? Solve CAPTCHAs? Scrape web pages? Verify token security? Fetch Polymarket odds? That's 318 APIs your agent doesn't know exist.

What you get

Every endpoint is probe-verified before listing. Dead endpoints are automatically deactivated. What you see in the catalog is what actually works.

How it works — the full flow

When Claude Code auto mode is active and the minia2a MCP server is connected, the flow for a paid API call is:

x-402-amount: 0
x-402-chain: base
x-402-token: USDC
x-402-register: POST /api/v1/register-simple

x-402-amount: 100
x-402-chain: base
x-402-facilitator: minia2a

Setup — 2 minutes

Add minia2a to your Claude Code MCP configuration. In your .claude/mcp.json or Claude config:

{
  "mcpServers": {
    "minia2a": {
      "transport": "streamable-http",
      "url": "https://minia2a.uk/mcp"
    }
  }
}

Restart Claude Code. The 318 tools appear automatically. No API keys, no .env files, no OAuth flows.

Verify it works

Test the MCP handshake directly:

# List all 318 tools
curl -s -X POST https://minia2a.uk/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}' | jq '.result.tools | length'
# → 318

# Call the free echo tool
curl -s -X POST https://minia2a.uk/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"minia2a_echo","arguments":{"message":"hello"}},"id":2}'

What this means for auto mode

Claude Code auto mode removes the human from the permission loop. But removing the human from discovery is the bigger unlock. An agent that can autonomously execute tools but doesn't know what tools exist is like a car with no map.

The minia2a MCP server is the map: 318 verified endpoints, each with a description, input schema, and price. The agent can:

• Search by capability: "I need to verify this token before swapping" → finds minia2a_token_security_check
• Compare prices: Multiple CAPTCHA solvers at different price points → picks the best fit
• Chain calls: Scrape a page → summarize the content → store the result — all with auto-pay

Why verified endpoints matter

The x402 ecosystem has an endpoint quality problem. An independent audit found that 76% of registered x402 endpoints return errors or time out. Many registries list endpoints that were deployed once and abandoned. The ERC-8004 registry has the same issue: agents self-declare x402Support: true but never enforce a paywall.

minia2a's approach is different: every endpoint in the catalog is probed before listing and re-probed periodically. Dead endpoints are automatically deactivated. The 318 tools in the MCP server are the ones that actually respond.

This matters for auto mode. An agent that tries 10 tools and gets 7 timeouts will learn not to trust the catalog. An agent where all 318 tools return real responses will integrate them deeply into its workflow.

What's next

The pieces are in place for Aug 14:

• ✅ Payment layer: x402 protocol with machine-readable 402 responses (amount, chain, token in headers)
• ✅ Discovery layer: 318 verified MCP tools with descriptions, schemas, and prices
• ✅ Credit system: Register → 500 free credits, no wallet setup needed
• ✅ Auto mode compatible: Agent can pay and retry without human intervention

The last mile is the .agent-budget standard — a file the agent reads to decide whether a paid call fits within its daily limit. But even without it, Claude Code auto mode + minia2a MCP is production-ready for trial-first discovery.

Get started: https://minia2a.uk/mcp · Catalog: /api/services · Docs: x402-help · Register: POST /api/v1/register-simple {"name":"my-agent"}