contact@ffxf.net
Console
API · MCP
API and MCP documentation
The console and the API take the same path: same quotas, same credit, same refusals. What the browser will not do, this will not do either, for the same reason. The MCP server gives an AI agent the same access, so it can rent a machine for the length of a task.
Get started
AI agents (MCP)
-
Connect an AI agent (MCP)
The FFxF MCP server in one sentence to give your agent, or one command for Claude Code, Codex and Cursor.
-
MCP tools
The sixty-eight tools of the MCP server, their arguments and the key scope each one needs.
-
Remote mode: the agent rents its machine
Let your agent rent a machine for a task, work on it over SSH and destroy it afterwards.
Understand
-
Authentication and scopes
API keys, scopes, address allowlists: what a key may do, and what it may not.
-
OIDC: CI without an API key
Call the API from GitHub Actions or GitLab CI with an OIDC token that lasts minutes, no stored key: provider, binding, conditions, refusals.
-
Credit, hours and budget
How hourly rental is paid, what an hour started means, and what stops a machine.
-
Private networks
Link your machines through an isolated network: create, order into it, connect without a restart, move.
-
Firewalls
Filter what comes into and out of your machines: reusable rules, enforced ahead of the machine, attached in one call.
-
Load balancers
Spread traffic across your machines: pools, listeners, routing by hostname and path, automatic certificates.
-
Managed databases
Managed PostgreSQL, MySQL, MariaDB and Valkey in your private network: TLS connections, users, point-in-time restore, forks, public access.
-
GitHub Actions runners
Your GitHub Actions jobs on a fresh machine in Montréal, paid by the minute: linking GitHub, labels, billing, tracking from the API.
-
Errors, limits and replay
The shape of a refusal, the stable codes, the rate buckets and idempotency.