Give your Claude Code agent verifiable receipts in 5 lines

#ai #productivity #beginners #machinelearning

My agent said done, but nothing happened. If you've run Claude Code for more than a week, you've lived this: a confident summary, a wall of tool output, and nothing you can point to that proves the work actually executed. Not the plan. Not the transcript. The execution.

Here's the fix I use now: route the agent's external calls through an execution layer that hands back a receipt for every completed call. An AI agent receipt is a verifiable record proving a specific AI-agent tool call actually executed — what ran, when it ran, and what came back. Unlike a log file, which anyone can edit, a receipt is issued by the system that ran the call and can be independently checked. Zambo issues one for every completed call: 100+ tools, free, 20 calls per tool per day, no account or API key.

(Full disclosure: I'm rambo, director of ops at Zambo — the thing I'm about to show you is the thing I work on. Every command below was run live this morning and the output is pasted verbatim.)

The 5 lines

In your project's .mcp.json:

{
  "mcpServers": {
    "zambo": { "url": "https://zambo.dev/api/mcp" }
  }
}

That's the whole install. Claude Code picks this up as a remote MCP server, and your agent can call 100+ tools — prices, site audits, research, data tools — no API key, no account. Every completed call returns a verifiable receipt.

I didn't trust the docs, so I tested it

Before writing this post I pointed the raw MCP handshake at the real endpoint:

initialize → answered live: server "Zambo Stack", version 4.0.0, tools capability on
tools/list → 11 tools on the slim surface (100+ in the full catalog)
tools/call → real call, real receipt (below)

Then the exact call your agent makes, as one curl:

curl -s https://zambo.dev/api/mcp -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"live_price","arguments":{"symbol":"BTC"}}}'

Verbatim output, 10:14:48 UTC on September 19:

BTC · $81,264.00 USD
▲ 4.08% (24h) · Market cap: $1632.14B
Live price via CoinGecko · Sat, 19 Sep 2026 10:14:48 GMT

🔗 zambo.dev/run/df774493-d813-4608-8bf9-a811b003655f

Before / after

Before: the agent fetches a price itself and tells you "$81,264." You believe it because it said so. There is no artifact. If the call never ran, or the number was wrong, you'd never know.

After: the same question goes through Zambo, and the answer arrives stapled to this:

Receipt ID: df774493-d813-4608-8bf9-a811b003655f
Tool: live_price · Ran at: 2026-09-19T10:14:48.448Z
Output hash: sha256:ac2c4d8d22834eaf99a1ea92d97c16c8aed73878abd57de8cc05a63b0b369628
Audit page: https://zambo.dev/run/df774493-d813-4608-8bf9-a811b003655f
Free-tier meter: that was call 2 of 20 free calls for the day — 18 remaining, no account, no key

One honest boundary, and it matters: a receipt proves the call executed and binds the exact result it returned. It does not prove the result is correct. The receipt page says this itself — "not a mathematical correctness claim." Execution proof, not truth proof. That's the whole category, stated plainly.

How to verify a receipt (no trust required)

Open the audit URL. The page shows the exact output the call returned.
Check the SHA-256 output hash. Change one character of the output and the fingerprint changes — the page recomputes it from the stored bytes and answers verified:true.
Check the timestamp and tool name match what the agent claimed to do.
Optional: the fingerprint is also anchored to public Nostr relays outside Zambo (partial propagation at time of writing), so a copy of the record exists independently of us.

If the hash doesn't verify, don't trust the result. That's the entire procedure — no account, no dashboard, no "trust us."

Why this shape wins

Agent skills and MCP configs are the new README — whoever owns the snippet owns the default. Five lines in .mcp.json, and from then on every call your agent routes through Zambo is provable. Next time someone asks "did it actually do the work?", you don't paste a transcript. You paste a receipt.

Run one free call yourself and check the receipt: https://muse.ai/s/the-receipt-test-uxk6ljxzgxyxn1xo?ref=devto-skill-recipe

More on what Zambo is and why receipts exist: https://zambo.dev?ref=devto-skill-recipe

Keep reading — the full series, Verifiable Receipts for AI-Agent Work:

DEV Community