MCP & WebMCP

Talk to Devhost the way you talk to your AI.

We expose our public surface as a Model Context Protocol server. Connect Claude, ChatGPT, Cursor, or any MCP-aware tool, and it can fetch our pricing, request a demo Odoo instance, or run a free security scan, without you copy-pasting between five tabs.

What this is

One paragraph.

MCP is an open protocol for connecting AI assistants to real services. We run an MCP server at https://devhost.io/api/mcp. Point your assistant at it and it can call our tools directly. You ask in plain English; the assistant invokes the tool; we return the answer. No SDK, no API key on the public-facing tools, no scheduled call.

Step one

Connect your AI assistant.

Pick the tool you already use. Each is a couple of lines.

Claude Desktop

Add Devhost to your MCP servers config

In Claude Desktop, open Settings → Developer → Edit Config. Add the entry below to your claude_desktop_config.json and restart Claude. The Devhost tools will show up in the tool picker.

{
  "mcpServers": {
    "devhost": {
      "url": "https://devhost.io/api/mcp"
    }
  }
}

OpenAI Responses API / ChatGPT

Pass our server as a tool

Models that speak the Responses API can attach a remote MCP server in one field. Same idea works in any platform that wraps Responses (Cursor, Vercel AI SDK, etc.).

import OpenAI from "openai";
const openai = new OpenAI();

const res = await openai.responses.create({
  model: "gpt-5",
  input: "What does Devhost charge for managed Odoo hosting?",
  tools: [{ type: "mcp", server_url: "https://devhost.io/api/mcp" }],
});

console.log(res.output_text);

Anything that speaks JSON-RPC

Or just curl it

Our MCP endpoint is plain HTTP + JSON-RPC. Useful for shell scripts and testing tools.

# List available tools
curl https://devhost.io/api/mcp \
  -H "Accept: application/json" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Or, in your browser

The page itself is a tool.

Every public devhost.io page calls navigator.modelContext.provideContext() with the same tools the MCP server exposes. So when you visit the site in a browser whose built-in AI agent supports WebMCP (an emerging Chrome and Edge feature), the agent can already call them, no setup, no extension.

Practically: ask the in-browser agent something like “ask Devhost for an Odoo 18 demo for me, my email is alex@example.com” and it fills in the right tool call with no further input from you.

If your browser doesn't have a built-in agent yet, the same tools are still available over MCP, see above.

What you can ask for

Tools, in plain English.

These are the same on the MCP server and in WebMCP. Schemas live in the agent-skills index.

get_pricing

Ask: What does Devhost charge for managed Odoo for a 30-user SaaS company?

Returns our current pricing for managed Odoo, custom engineering, security audits, DevOps retainers. The agent can do the math; you don't fill out a form.

request_odoo_demo

Ask: Spin up an Odoo 18 demo for me, send it to alex@example.com.

We provision a personal sandbox with sample data. Live for a few days. The agent collects the version + email, calls the tool, you check your inbox.

run_security_assessment

Ask: Run a free security scan on example.com and email the report to me@example.com.

Non-invasive scan of DNS, TLS, exposed services, known vulns. Short report by email. No credit card.

talk_to_odoo_agent

Ask: Open the Odoo demo agent so I can poke at the data.

Hands-off chat against a sample Odoo instance. The agent can both answer questions about the data and execute actions inside Odoo.

Skip the sales call.

Every public capability we have is already wired to your AI assistant. Use it, then come find Aric if you want to talk to a human.