Four open ledgers.
Marketing makes claims. Code keeps records. Below: the four ledgers that make tools402 verifiable in real time — facilitator state, on-chain settlements, the install command itself, and the agent-native toolkit. Pulled from production, refreshed while you read.
The facilitator ledger.
tools402 routes settlements through Coinbase CDP first, then Cloudflare, then PayAI. When the primary goes silent — like January 2026 — the chain rolls forward in under 500 ms. State below is live, refreshed every 60 seconds.
The settlement ledger.
Each row is a real call by a real agent, paid in real USDC on Base mainnet. The transaction hash links to Basescan — verify any one of them. We hold no funds; the agent settles directly to a public recipient wallet.
0x3b9F…759e
The install ledger.
If your agent runs in Claude Desktop, Cursor, or Cline, you're three lines of JSON away. Paste the snippet for your client into its config file — your agent gains 156 paid utility endpoints, paid in USDC, no account required.
~/Library/Application Support/Claude/claude_desktop_config.json// merge into your existing mcpServers { "mcpServers": { "tools402": { "command": "npx", "args": ["-y", "@tools402/mcp"], "env": { "TOOLS402_SESSION_TOKEN": "<your-token>" } } } }
~/.cursor/mcp.json{
"mcpServers": {
"tools402": {
"command": "npx",
"args": ["-y", "@tools402/mcp"],
"env": { "TOOLS402_SESSION_TOKEN": "<your-token>" }
}
}
}
cline.mcpServers{
"cline.mcpServers": {
"tools402": {
"command": "npx",
"args": ["-y", "@tools402/mcp"],
"env": { "TOOLS402_SESSION_TOKEN": "<your-token>" }
}
}
}
SESSION=$(npx tools402-authorize --amount 10 --days 30) && jq --arg t "$SESSION" '.mcpServers.tools402 = {command:"npx",args:["-y","@tools402/mcp"],env:{TOOLS402_SESSION_TOKEN:$t}}' ~/Library/Application\ Support/Claude/claude_desktop_config.json > /tmp/c.json && mv /tmp/c.json ~/Library/Application\ Support/Claude/claude_desktop_config.json
The agent toolkit ledger.
Five primitives that agents actually need — memory, sandbox, browse, critique, judge — built on the same x402 rails as the rest of the catalog. Paid by the call, or bundled as prepaid agent kits. No SDK lock-in, no platform tax beyond the atomic USDC price.
POST /v1/agent/memory/upsert $0.0005/item Pinecone vector memory · semantic + metadata POST /v1/agent/memory/query $0.001/query top-k similarity search across namespaced pools POST /v1/agent/memory/delete $0.0001/item delete by id list, namespace-scoped POST /v1/agent/sandbox $0.01/60s python · node · bash on Modal · network access toggle POST /v1/agent/browse $0.002/page Playwright screenshot + DOM · isolated worker · egress filter POST /v1/agent/critique $0.05/call 3-model critique (Claude + GPT + Gemini) + synthesis POST /v1/agent/judge $0.08/judgment 5-model consensus vote · majority/weighted/unanimous
POST /v1/_bundles/research-agent/purchase $0.50 200× url-text + 100× sitemap-parse + 40× nlp/sentiment POST /v1/_bundles/document-agent/purchase $0.30 50× ocr + 20× md-pdf + 15× memory/store POST /v1/_bundles/voice-agent/purchase $1.00 100× whisper + 3× voice-clone + 1× lipsync // returns a session token tb_<hex> — pass it via X-Bundle-Token header // instead of X-Payment, server deducts from the credit pool. 30-day expiry.
Want the proof to be wrong?
Run it. The cheapest paid call is 0.0001 USDC. Replay-protected, on-chain, no account. If anything above is theatrical, you'll know in 30 seconds.