Request Early Access Getting Started Guide

Build and run AI agents with Claude. No code. No infrastructure. Just describe what you want and agents come to life.

How it works

Step 1: Connect to the MCProspero MCP server.

Add MCProspero to Claude Desktop, claude.ai, or Claude mobile. One-time setup, takes 30 seconds. (Actually any MCP client, like ChatGPT, will work — we just really like Claude)

Step 2: Connect your tools.

Link your Gmail, Outlook, Slack, Calendar, GitHub — whatever your agents will need. Revoke access anytime.

Step 3: Tell Claude what you want.

"Check Hacker News every morning and email me a summary of the top AI stories." Claude designs the agent, writes the prompt, picks the tools.

Step 4: Review and approve.

MCProspero shows you exactly what the agent will do — which tools it uses, who it contacts, information sources it accesses. But the agent does not start until you approve the plan.

Step 5: Monitor and modify.

Check on your agents anytime, from any device. Modify with a follow-up conversation. "Forward the HN summary to my boss too." MCProspero handles the change.

Build agents by talking

Create agents from your desktop, web browser, and mobile phone.

Claude with MCProspero
Check my email every hour for unread messages from my wife. If anything is urgent, summarize and DM me in Slack.
C

Let me set up an agent for this.

⚙ list_available_tools
⚙ list_connections

Gmail and Slack are both connected.

C
⚙ create_agent  ✓

Here's what the agent will do:

Read Email: Personal Gmail
Send Slack: Direct message to you

Approve?

Yes
C

Urgent Email Watch is live. It checks your Gmail hourly for urgent messages from your wife and DMs you a summary in Slack.

Running

Judgment and safety

AI agents can make judgment calls — "Is this email urgent?" — but they need guardrails.

Manifest approval

Before an agent runs, you see everything it plans to do — every tool, every contact, every information source. You approve the list. At runtime, anything not on the list is blocked and the agent is halted.

Secrets scanning

All data flowing through agents is scanned for API keys, tokens, and passwords in real time. If an agent tries to send a secret, it's stopped immediately.

PII protection

Phone numbers, credit cards, SSNs — detected automatically in every tool call. You choose what happens: ignore, warn you, or halt the agent. Per-action overrides let you suppress scanning where you expect PII to be present.

No vendor lock-in

Any MCP client. Any OpenAI-compatible model — Claude, OpenAI, Llama, Mistral, or whatever comes next. Open standards on both sides.

Easy enough for your boss — safe enough for your kids (and your CISO).