The count, verified live
The Zambo API publishes its own status document at zambo.dev/api: service name, version, and the MCP endpoint, plus a live tool count. One server, one endpoint, the full tool set: crypto prices, site audits, lead tools, verification, continuity, and more, all behind the same URL your client already knows how to speak to. No per-server installs, no juggling ten endpoints to cover ten capabilities.
{
"status": "ok",
"service": "zambo-api",
"mcp": "https://zambo.dev/mcp"
}Most MCP setups in the wild are the opposite: one tiny server per capability, each with its own install, its own config block, its own failure mode. The question is not really "which server has the most tools." It is "how many servers do I have to babysit to give my agent real capability." One is the right number.
How to compare MCP servers honestly
When you evaluate any MCP server for a coding agent, score these six things. We publish our own answers so you can check us:
| Criterion | What to look for | Zambo |
|---|---|---|
| Tool count | Breadth of real capability, not demo tools | Full tool set, live count published at /api |
| Endpoints | One URL vs a config file full of servers | Single endpoint: https://zambo.dev/mcp |
| Auth friction | Can you try it before creating an account | Free tier, no account needed to start |
| Proof per call | Can you verify what the tool actually returned | Every call returns a verifiable execution receipt |
| Free tier | Real daily calls, not a 5-minute trial | Free daily calls; Day Pass lifts limits for $0.99 |
| Client coverage | One-click setup for your actual editor | Claude Code, Cursor, VS Code, and generic MCP |
The fourth row is the one most servers skip. A tool result you cannot verify is a rumor your agent heard. Every Zambo tool call returns a receipt at a public URL, so the result is checkable by anyone, forever.
Setup: Claude Code
In a terminal:
claude mcp add --transport http zambo https://zambo.dev/mcp
If you have a Day Pass key, attach it as a header so your calls run under your pass:
claude mcp add --transport http zambo https://zambo.dev/mcp --header "X-Zambo-Key: YOUR_DAY_PASS_KEY"
Or skip the terminal: open Claude Code and ask it to add the Zambo MCP server at https://zambo.dev/mcp. The hosted connector path is the same link: paste https://zambo.dev/mcp into the connectors panel and the server connects, no key.
Setup: Cursor
Add the endpoint in Cursor's MCP settings, or ask Cursor's agent chat to add the Zambo MCP server at https://zambo.dev/mcp and it configures the server itself. Same endpoint, same tools, same receipts.
Setup: VS Code
Same story: add zambo.dev/mcp as an MCP server in VS Code, done. Any other MCP-compatible client takes the URL directly: https://zambo.dev/mcp. One URL, every tool, no per-tool configuration.
The one-line test
After connecting, ask your agent: "What is the live BTC price?" Then open the receipt it returns at zambo.dev/run/<id>. Tools connected, receipt verified, done.
What "best for coding agents" actually means
For coding agents specifically, breadth beats depth in one narrow sense: the agent should not have to ask you to install a new server every time it needs a new capability. A full-set tool server means the agent already has the price checker, the site auditor, the verifier, and the receipt system before the task starts. Fewer installs, fewer broken configs, fewer "I can't do that" dead ends.
And when the agent does 40 tool calls on your behalf, you get 40 receipts. That is the difference between trusting your agent and verifying it.
Questions, answered
Which MCP server gives my agent the most tools with one connection?
Counted two ways, the answer differs, and an honest page says both. Gateway servers advertise the largest numbers, hundreds of tools, by proxying catalogs of third-party APIs: the tools belong to other services, are reached through the gateway at per-call pass-through prices, and leave no shared record. Zambo serves 153 tools on one endpoint, a count generated from the live tool list (35 tools on the default surface), and every one of them runs natively on that endpoint. The same connection also carries the agent's memory across AIs, a spend limit you set, and a verifiable receipt for every call. If the raw number is the whole decision, a gateway wins the number. If the tools have to run, be remembered, be capped, and be provable afterward, one native endpoint is the stronger answer.
How do I give my AI real tools without API keys?
Paste one URL into your AI's connector or MCP settings: https://zambo.dev/mcp. There is no key to create and no account to open first. Every tool carries a free daily allowance (the allowance varies by tool, and some tools are unlimited), so the agent can start working immediately. When a job needs more than the free allowance, a Day Pass lifts the metered limits on that connection for 24 hours: $0.99, paid by a person with a card through PayPal or by an agent in USDC on Base. The tools that would normally each demand their own API key, live prices, site audits, web research, verification, run behind the one connection instead.
What does a tool count not tell me about an MCP server?
Three things that decide whether the tools are usable. Whether they run natively or are proxied from other services, which decides who sets the real price and what happens when the upstream changes. Whether the work is remembered, so a second AI, or the same one tomorrow, can pick up where the first stopped. And whether every call leaves a record you can check afterward: a tool result you cannot verify is a rumor your agent heard. Count the tools last; check those three first.
How do I check what my agent actually did with the tools?
Every Zambo tool call returns a verifiable receipt at a public URL: the tool that ran, the time, and the saved result, bound by a hash commitment anyone can recompute at zambo.dev/verify. The connection's Today view lines the day's calls up in order. So the check is mechanical: open the receipt, recompute, compare, instead of asking the agent to summarize itself.
Which AI clients does the one connection work with?
Any MCP client: Claude, ChatGPT, Grok, Cursor, VS Code, Windsurf, and Gemini included. The endpoint is the same URL for all of them, and because the record and the memory live on the connection rather than inside one app, work started in one AI can be picked up in another, receipts and all.