1. Claude Code
Save this exact JSON as .mcp.json in your project root. Then restart Claude Code and approve the server when prompted.
{
"mcpServers": {
"zambo": {
"type": "http",
"url": "https://zambo.dev/mcp"
}
}
}
The hosted MCP server for agent work
Give your AI one connection to get real work done.
20 calls per tool per day, no account
20 free calls per metered tool each day, no account, no key
Connect to one hosted MCP server with 138 live tools. No API key or account required to get started. Every call produces a verifiable receipt, so the result can be inspected after the work is done.
https://zambo.dev/mcpConnect in minutes
Add the endpoint to your MCP client. Start with the Claude Code configuration, or use the same URL with any client that supports remote Streamable HTTP MCP servers.
2. Any MCP client
Create an HTTP/remote MCP server connection named zambo and paste this endpoint. No environment variables or credentials are required.
{
"name": "zambo",
"type": "http",
"url": "https://zambo.dev/mcp"
}
The endpoint uses the standard MCP HTTP transport. Keep the same endpoint for normal MCP discovery and tool calls.
Session lifecycle: use new_session to start a fresh session. Passing the returned session ID continues that session; changing or dropping the ID starts clean context.
Evidence for every call
Every call proves itself.
Tool output is paired with a verifiable receipt that records what happened. Inspect a real example where a prompt-injection attempt was blocked and the run received a risk score of 95/100.
Inspect the verifiable receiptA result you can check
Open the public receipt to review the recorded run, its outcome, and the risk assessment without relying on an unverified claim in this page.
Also listed on
Find Zambo's server listing and published service profiles on these live directories.
Ready to get connected?
Choose the setup that fits your client and make your first call.
See all installation options