Kromatic MCP servers for your AI assistant
Kromatic runs a small estate of public Model Context Protocol (MCP) servers, each reachable from the AI chat assistant you already use. Compare them below, connect in a couple of minutes, and see how Fermi and Neumann compose into one workflow.
Compare the servers
| Server | What it does | Free tier | Connect |
|---|---|---|---|
| Fermi | Calibrated probability estimates for any quantitative question. | Free anonymous tier: 3 estimates per UTC day, no sign-up needed — or create an account for 3 free credits. | https://fermi.kromatic.com/mcp |
| Neumann | Monte Carlo simulation for plans with multiple uncertain inputs. | Free anonymous tier: 3 runs per UTC day, up to 1,000 trials and 20 nodes per run — no sign-up needed. | https://neumann.kromatic.com/mcp |
| Krobot | Krobot gives AI assistants written guides that product managers, founders and teams follow to test demand, win first paying commitments and decide whether to keep funding. | Free; no sign-in to browse, search and read — a free Kromatic account (no card, no approval) to download templates and send feedback; paid courses and templates are bought on kromatic.com. | https://api.kromatic.com/mcp/krobot |
Connect in your assistant
Pick a server below to see its install snippet for your client.
- Open Claude.ai and go to Customize → Connectors, then choose "+ Add" → "Add custom connector".
- Name it "Fermi" and paste this server URL: https://fermi.kromatic.com/mcp
- Click Add, then turn it on in any chat from "+" → Connectors. On Team or Enterprise, an Owner adds it first under Organization settings → Connectors.
https://fermi.kromatic.com/mcpclaude mcp add --transport http fermi https://fermi.kromatic.com/mcp- Open ChatGPT on the web and go to Plugins (chatgpt.com/plugins), then choose "+" → "Add custom MCP server".
- Name it "Fermi" and paste this server URL: https://fermi.kromatic.com/mcp
- Accept the warning, choose "Create as a plugin" and install it, then type "@Fermi" in a chat to use it. Workspace admins may need to allow custom MCP servers first.
https://fermi.kromatic.com/mcp- Go to grok.com/connectors, click New Connector, then select Custom.
- Enter this MCP server URL: https://fermi.kromatic.com/mcp
- Grok discovers Fermi's tools and uses them in your conversations. On Grok Business or Enterprise, a team admin must provision the connector first.
https://fermi.kromatic.com/mcp{
"mcpServers": {
"fermi": {
"url": "https://fermi.kromatic.com/mcp"
}
}
}{
"servers": {
"fermi": {
"type": "http",
"url": "https://fermi.kromatic.com/mcp"
}
}
}{
"mcpServers": {
"fermi": {
"httpUrl": "https://fermi.kromatic.com/mcp"
}
}
}Compose them: Fermi → Neumann
Fermi and Neumann are built to be used together: ask Fermi for a calibrated range on each uncertain input, then hand those ranges to Neumann to see what they imply once they interact.
- Ask Fermi for ranges. Get a calibrated 90% interval for each uncertain input instead of guessing a single number.
Using the Fermi MCP server, give me a 90% interval for our monthly conversion rate (visitors to paying customers), and a separate 90% interval for our price per unit. - Hand those ranges to Neumann. Turn Fermi's two intervals into distributions and simulate revenue over 1,000 trials.
Using the Neumann MCP server, model revenue = conversion_rate × traffic × price. Turn the conversion-rate and price intervals Fermi just gave you into distributions for conversion_rate and price, and run 1,000 trials. - Read the results. Look at the spread, not just the middle, and see which input to de-risk first.
From that Neumann run, report the p10, p50 and p90 of revenue, and tell me which input has the largest sensitivity so I know what to de-risk first.
Learn more
Contact our experts
