Point it at the app
Paste a URL. MCPer.io finds the OpenAPI or REST surface and reads it. No API? Record the flow once in the built-in browser recorder, or let an AI operator drive the UI.
Hosted MCP servers for any app
Turn any app into an MCP server. We host it.
Point MCPer.io at your API or web app. It scans every endpoint, describes each one with AI, switches destructive operations off until you say otherwise, and serves the result from yourapp.mcper.io. No server to write, deploy or keep alive. Claude, ChatGPT, Cursor and every other agent connect in minutes.
Works withClaudeClaude CodeChatGPTCursorWindsurfany MCP client
https://warehouse.acme.comScan
- Found OpenAPI 3.1 at /openapi.json · 23 endpoints
- Described all 23 with AI, in the words an agent needs
- Sign-in detected: API key per user. Stored encrypted.
- 4 destructive operations switched off by default. Turn any on when you're ready.
- Writes preview first and need confirm: true
warehouse.items_listread
warehouse.movement_createwrite · confirm
warehouse.item_deletedestructive
live · mcp://acme.mcper.io/warehouse
How it works
Scan. Shape. Ship.
Three steps from a URL to a connector your whole team's agents can use. No server to write, no SDK to learn, nothing to redeploy when the app changes.
Decide what agents may do
Every endpoint arrives described by AI and sorted by kind. Keep the reads, gate the writes behind a preview, leave the deletes off. Presets lock the fields an agent must never get wrong.
We host it. You share a URL.
The connector goes live at yourapp.mcper.io on our infrastructure: TLS, auth, scaling and updates included. Your team adds one MCP URL to Claude, ChatGPT or Cursor and signs in as themselves.
Nothing to host
The MCP server you never have to run.
Most MCP servers die in a repo nobody maintains: a process to deploy, certificates to renew, auth to bolt on, an on-call rota for a thing that was supposed to save time. MCPer.io hosts the endpoint for every app you connect, so the only thing you ship is a URL.
Build and host your own MCP server
- ✕Write a server per app, keep it in sync with the API
- ✕Deploy it, expose it on the internet, renew TLS
- ✕Bolt on auth, per-user identity and secrets storage
- ✕Find out a tool broke when an agent does something wrong
- ✕Repeat for the next app, and the one after
Connect it to MCPer.io
- ✓Scan the app, review the tools, press publish
- ✓Live at yourapp.mcper.io/mcp with TLS, rate limits and uptime handled
- ✓SSO, per-user credentials and an encrypted vault built in
- ✓Nightly tests quarantine a broken tool before anyone hits it
- ✓Every app, every change, from one portal. No redeploys.
Need it inside your own network? Enterprise plans can run MCPer.io in a dedicated region or self-hosted, with the same portal.
What's inside
Built for apps that were never designed for agents
Most software has no MCP server and never will. MCPer.io meets each app where it is: an API if there is one, a browser if there isn't, and the same safety rules either way.
API from OpenAPI
Import a spec or let the scan find it. Tools are generated from the endpoints, named for humans and described for agents. The gold standard when it exists.
Scripted browser
No API and a stable UI? Click through the flow once in the recorder. MCPer.io replays it with fixed selectors, screenshots every step, and marks the preview point.
AI browser
For UIs that keep changing. Describe the goal, allow-list the URLs, and an AI operator reads the screen and acts. The final submit is always gated.
Presets lock the risky fields
Project, queue, reporter, priority rules, naming templates: set once by the owner, enforced on the server. The agent never gets to improvise them.
Nightly tests, automatic quarantine
Every connector runs against a sandbox each night. A failed step flips the app to degraded and hides the tool from agents, with a screenshot diff and a suggested fix waiting for you.
Fully hosted, always on
Every connector runs on MCPer.io's infrastructure at your own subdomain. TLS, scaling, rate limits and protocol updates are ours to worry about; you never deploy a thing.
Every call runs as the person
Developers sign in with your SSO and connect their own credentials, encrypted at rest. Audit shows who asked for what, and what the app actually did.
Safe by default
Agents move fast. Your data shouldn't break.
- Destructive operations are off until you turn them on. DELETE, purge, reset and anything the AI flags as irreversible start disabled. Agents never see them.
- Writes preview first. The agent shows the person exactly what will be sent. Only confirm: true with a signed token submits it.
- Allow-lists, not hope. Browser tools can only visit the URLs you approve. Credentials live in an encrypted vault; nothing is logged in clear.
- An audit trail you can read. Every call, every preview, every confirmation, with screenshots wherever a browser was involved.
Acme Helpdesk · tools agents can see
helpdesk.tickets_list
List the caller's tickets with status.
helpdesk.ticket_create
Open a ticket. Previews, then needs confirm.
helpdesk.ticket_comment
Reply on a ticket the caller owns.
helpdesk.ticket_delete
DELETE /tickets/{id}. Off by default.
helpdesk.purge_closed
POST /tickets/purge. Off by default.
Agent: I'll open a Network access ticket, priority High, for Lev. Submit?
→ confirm: true · token c7f1… · created #2314
→ confirm: true · token c7f1… · created #2314
Any MCP client
One URL. Every agent your team already uses.
MCPer.io speaks standard Model Context Protocol over Streamable HTTP. Add the endpoint once; each person signs in as themselves and sees only the apps your workspace has enabled.
Claude DesktopClaude CodeChatGPTCursorWindsurfVS CodeAgentDeckyour own agent
// claude_desktop_config.json · cursor mcp.json · any MCP client
{
"mcpServers": {
"acme": {
"type": "http",
"url": "https://portal.mcper.io/mcp/k_7f3a9c"
}
}
}
Who it's for
Internal tools, products, and everything in between
Give every developer's agent the company's Jira, HR portal, ERP and that one internal form, with the organisation's rules baked in. Attach apps from a dashboard; never write a connector again.
Your customers are asking "do you have an MCP server?" Point MCPer.io at your API and ship one this week, with safe defaults you can stand behind.
Connect a client's stack in an afternoon, hand them the portal, and keep the nightly tests watching after you leave.
Pricing
Start free. Pay when agents start working.
Free
$0 / month
- 1 app, up to 5 tools
- API connectors
- Preview and confirm on writes
- Community support
Team
$49 / month
- Unlimited apps and tools
- API, scripted browser and AI browser
- Nightly tests with suggested fixes
- SSO, per-user credentials, audit
- Email support
Enterprise
Custom
- Self-hosted or dedicated region
- Bring your own model key or local agent
- SAML, SCIM, custom retention
- Dedicated support and SLA
Prices shown are indicative for launch and may change.
FAQ
Questions people ask before connecting
What is an MCP server, and why would my app need one?
The Model Context Protocol (MCP) is the open standard that lets AI agents such as Claude, ChatGPT and Cursor call tools. An MCP server exposes your app's capabilities as those tools. Without one, agents can only talk about your app; with one, they can act in it on a person's behalf.
Do I have to host or deploy anything?
No. MCPer.io hosts the MCP endpoint for every app you connect, at your own subdomain, with TLS, authentication, scaling and protocol updates handled. You publish from the portal and share a URL. Enterprise plans can additionally run in a dedicated region or self-hosted.
Does my app need an OpenAPI spec?
No. If a spec exists, MCPer.io finds and imports it. If there is a REST API without a spec, it probes and lets you describe endpoints. If there is no API at all, you record the browser flow once, or let the AI operator drive the UI inside an allow-list.
What happens to delete and other destructive endpoints?
They are switched off by default and never shown to agents. You can turn any of them on per tool, and even then it previews first and needs an explicit, signed confirmation.
Do agents act as a shared service account or as the person?
As the person. Developers sign in with your SSO (Google, Microsoft or any OIDC provider) and connect their own credentials, which are encrypted at rest. Every call and preview is audited under their name.
Which agents and clients does MCPer.io work with?
Anything that speaks MCP over Streamable HTTP: Claude Desktop, Claude Code, ChatGPT, Cursor, Windsurf, VS Code and custom agents. You add one connector link; the agent signs you in with Google and you pick the tools it may use.
What if the app's UI changes and a browser tool breaks?
Nightly tests catch it. The broken tool is hidden from agents within a minute, and the portal shows you the baseline and tonight's screenshots side by side with a suggested fix you can apply and re-run in the sandbox.
Get started
Your app, speaking MCP, before lunch.
Connect the first app free. No credit card, no SDK, nothing to host. Destructive operations stay off until you decide.