Your systems, callable by any AI agent.
You decide which tools an AI may use: look up a customer, query the warehouse, file a ticket. foro runs them live in a minute, hosted in Europe. Every call ends up on the record.
Free tier: 10,000 tool calls a month, no credit card, and your project stays as it is.
A diagram of four tool calls, left to right. An AI client asks a question in plain language, it reaches foro's gate where it is authenticated and isolated, runs against one of your systems as a named tool call, and the answer returns the same way, masked for PII and logged as it passes back through the gate. “Which accounts churned?” becomes crm.search_accounts against your CRM, returns 12 accounts, and comes back as “Twelve, mostly on annual plans.”. “Revenue by region, Q3?” becomes warehouse.run_query against your warehouse, returns 1,204 rows, and comes back as “EMEA up, APAC flat.”. “File a bug for this crash” becomes tickets.create against your tracker, returns TICKET-4417, and comes back as “Filed, and on the on-call queue.”. “Find our refund policy” becomes docs.search against your docs, returns 8 documents, and comes back as “Thirty days, unopened only.”. “Buy 3 more of these” becomes store.create_order against your storefront, returns order #8821, and comes back as “Ordered, arrives Thursday.”.
What teams build MCP servers for
An MCP server is a small piece of software that lets an AI agent use one of your systems. Teams build them for two reasons. foro is built for both.
For your customers
Show up inside the AI your customers already use
Your customers spend their day in Claude, Cursor and Le Chat. An MCP server is how your product appears there, beside everything else they use.
- A read-only wrapper over the API you already publish
- Your docs, answerable in the agent's own window
- The tasks customers log in for: check a status, update a record
Every deploy writes the snippet your users paste into their client. With MCP Apps, it can bring its own interface too.
For your own team
Power your agentic workflows by connecting your source systems
Your CRM, your data warehouse, your ticketing, your documents: each one needs a server that exposes only the actions you pick. Run one per system, each sealed off from the others, each on its own token.
- Your warehouse, queryable by an analyst's agent
- CRM and ticketing, so an agent reads the case and drafts the reply
- Internal APIs and runbooks, callable by the agents your engineers run
One secure environment and one private network per system. A leak in one reaches nothing else.
Writing an MCP server is the easy part. Deploying one safely is the rest of the week.
Do it yourself, on AWS, GCP or Azure
0 / 14Build
Route
Secure
Operate
On foro
- Sign in
- Pick your project
- Click Deploy
- Live
Live in a minute. What every server gets:
- A URL that never changes
A stable
https://<slug>.foro.sh, or a domain you own. - Locked before it is live
A bearer token in front of every deploy, until you switch it to OAuth.
- Personal details out of the payload
Emails, IBANs, names and more masked in a result before it reaches the AI provider.
- Every call on the record, per tool
Calls, p95, errors and token cost, per tool, with live logs alongside.
- A deploy that cannot take you down
Health-checked; the old container serves until the new one passes.
Then watch it run
Every deployed server comes with a dashboard: calls, latency, tokens, and errors per tool, request sources, uptime, how much PII the gate masked out of tool results, and live logs. Push a fix and redeploy in one click. Nothing to instrument in your code.
5 projects · 4 live · 1 building
Calls · 7d
Calls · 7d
Calls · 7d
Calls · 7d
Your server can return an interface, not just text
An MCP App is the server you already have, plus one resource that is a view, and the tools that drive it. The client renders that view in a sandboxed frame, your buttons call your tools back, and your product arrives in the conversation wearing its own interface.
ui:// resources and the app config are part of the MCP spec. foro deploys a server that declares one exactly like any other, with nothing to turn on. How it works → Why it matters →
Todo list
The todo template, as a client renders it.
The two questions a security review usually gets to
"Show me what the agent actually did."
You get a record of every call, per tool: what ran, how long it took, whether it errored, with the live logs next to it. From the Starter plan, emails, IBANs, card numbers and phone numbers are masked on the way out, before the response leaves your server. How it works →
"Where does the data sit, and who can compel it?"
Your server runs on OVH in France. Nobody in that ownership chain sits outside the EU, so there is no company holding your data that a US CLOUD Act order can reach. Your source, your images and your logs fall under GDPR and the EU Data Act. How it works →
Some data still leaves the EU, and it does so on paths you choose: your GitHub repo, your own model provider, a package registry. The privacy policy names every one.
Pick a plan. Live in a minute.
Every plan runs the same platform. What changes is capacity and support.
Free
no card required
€0/mo
- 1 MCP server
- 1 seat
- 10,000 tool calls / month
- 24-hour logs, 7-day metrics
- Community support
Starter
for solo developers in production
€19.00/mo
VAT included
first month €9.50
- 3 MCP servers
- Auto-deploy on commit
- Structured PII masking
- Email support
Team
for teams and consultancies
€249.00/mo
VAT included
- 10 MCP servers
- 10 seats, workspaces & roles
- Structured + free-text PII masking
- Priority email support
Enterprise
for organisations that need more
Custom
- Custom limits & seats
- Custom DPA
- Guided onboarding, including SSO with your own IdP
Billed monthly. Cancel any time; your plan runs to the end of the month you have paid for.
Frequently asked questions
You can, but that means a cloud with a US parent even in its "EU region", the DevOps time to isolate and TLS-front a container safely, the risk of a rushed first deploy leaking a secret, and building your own metrics and logs from nothing. Learn more →
Any client that speaks streamable HTTP. The dashboard has ready-made setup for Claude Desktop, Cursor, Codex, VS Code, Windsurf, and Antigravity (a copy-paste config snippet) plus Mistral Le Chat (a URL and token for its custom connector). Learn more →
No. It's the same server, plus one resource that is a view and the tools that point at it. Nothing to enable and no separate deploy. Learn more →
Yes. Most MCP clients let you add more than one server, and an agent can call tools across all of them in the same conversation. Each server still runs isolated, on its own token. Learn more →
Every deploy is health-checked before it goes live, and redeploys swap in the new container only once it passes. The previous one keeps serving traffic until then, so a bad deploy never takes your server down. Learn more →
Every server runs on European cloud infrastructure, on every plan, not just Enterprise. No non-EU parent company sits in the ownership chain, and logs and backups are stored in the EU. The one non-EU touchpoint is GitHub, when you choose to deploy from a GitHub repo. Learn more →
Your first server, live in a minute.
Sign in, pick your project, click Deploy. No credit card required.