Servidor MCP de FFxF

• Nome no registronet.ffxf/cloudverificado no registro MCP

O nome no registro fica sob ffxf.net. O registro só libera esse espaço de nomes para quem prova controle do domínio por DNS ou HTTP.

• Estadorespondeverificado em 2026-10-29
• Ferramentas71
• O servidor se chamaffxf
• Versão do servidor1.12.0
• Versão do protocolo2025-06-18

Em 2026-10-29 o servidor respondeu a initialize e tools/list sem login. Primeira resposta em 234 ms.

• Ainda não avaliamos se as ferramentas fazem o trabalho do produto.
• Verificado em 6 dos 7 dias necessários para julgar a confiabilidade.

Dito por regras a partir das medições acima. A primeira chamada provável é a escolha de um modelo na lista de ferramentas; leia a descrição antes de confiar nela.

• Nota INITE MCPdepois de uma semanaa nota precisa de 7 verificações diárias: 6 até agora
• Respondeumedindo6 das 7 verificações diárias necessárias para mostrar a porcentagem
• Mediana da primeira resposta289 ms
• Espaço no contexto do agente≈ 12.835 tokensas definições de todas as ferramentas (71), que o agente carrega a cada turno
• Definições das ferramentasA · 94 de 100
• Resultados tipados0 de 71 ferramentasdeclaram outputSchema: o resultado chega tipado e é lido de uma vez

• Parâmetros sem descrição: em 27 de 71

Um programa verifica o servidor uma vez por dia: initialize e tools/list, sem modelo. Os tokens são o tamanho das definições, a cerca de quatro caracteres por token. A nota é a média de seis verificações do nome, da descrição, do esquema e das anotações de cada ferramenta; diz se as ferramentas estão bem descritas, não se funcionam. A nota pesa só o que foi medido: se responde (25%), se faz o trabalho do produto (25%), definições (20%), contexto (10%), uso (10%) e a facilidade de conectar (10%). Uma parte não medida fica de fora, e uma instrução oculta limita a nota a 40. O estado dos servidores MCP dos fornecedores: método e números →

• list_regions

  Where machines can run. Public, no key needed.

• list_plans

  Plans with their resources and prices (hourly and monthly, in CAD). Public, no key needed.

• list_images

  Operating systems and ready-to-use application images. The slug is what create_vm expects. Public, no key needed.

• get_account

  Credit balance, hourly burn rate and runway, quota and budget caps. Scope: billing.read.

• list_vms

  The account's machines with their status, addresses and billing. Scope: vms.read.

• get_vm

  One machine: status, IPv4, IPv6 /64, specs, traffic (Internet only, this traffic month: used_gb, limit_gb, period_ends_at), billing, suspension. Scope: vms.read.

• get_vm_metrics

  CPU, memory, disk and network as measured from outside the machine. Scope: vms.read.

• estimate_vm

  Runs every check of create_vm and prices it, without creating anything: would_succeed, hourly rate, monthly price, credit required. Call it before create_vm. Scope: vms.create.

• create_vm

  Orders a machine and spends credit (hourly) or raises an invoice (monthly). Returns the machine and a provisioning action to poll with get_action. Retrying the same order is safe: it is deduplicated for 24 hours. Scope: vms.create.

• power_vm

  start, shutdown (clean ACPI), reboot, or stop (hard power-off). Returns an action to poll with get_action. Scope: vms.action.

• rename_vm

  Changes the hostname, inside the guest and in the console. Scope: vms.action.

• reinstall_vm

  ERASES THE DISK and redeploys another image. Ask the user first. confirm_hostname must be the machine's exact current hostname. Scope: vms.action.

• destroy_vm

  DESTROYS the machine and its disk, with no undo and no backup kept. Ask the user first. confirm_hostname must be the machine's exact hostname. Scope: vms.destroy.

• get_action

  Follows an asynchronous action (create, reinstall, power…). Poll until status leaves running; on error it carries the reason. Scope: vms.read.

• list_vm_actions

  History of actions on one machine, from the API and the console. Scope: vms.read.

• get_vm_network

  IPv4 and IPv6 addresses of a machine with their reverse DNS, and its private networks. Scope: vms.read.

• set_reverse_dns

  Sets the PTR record of one of the machine's addresses; null restores the default. Scope: network.write.

• list_vpcs

  The account's private networks with their range, gateway and connected machines, plus the quota. Scope: vms.read.

• get_vpc

  One private network and the machines on it, with their private address. Scope: vms.read.

• create_vpc

  Creates an isolated private network; ready when the call returns. The first usable address (.1) is reserved for the gateway. Scope: network.write.

• delete_vpc

  Deletes an empty private network. Refused while a machine is still on it: disconnect or destroy those first. Ask the user first. Scope: network.write.

• connect_vpc

  Hot-adds a card on the private network and sets its address in the system, without a restart. configuration: manual means the address must be set by hand in the guest. Connecting again returns the same address. Scope: network.write.

• disconnect_vpc

  Removes the machine's private address and card, without a restart. Refused for the only network of a machine with no public address. Scope: network.write.

• move_vpc

  Joins the target network, then leaves this one, so the machine is never cut off. Keeps the same address ending when free. Scope: network.write.

• list_firewalls

  The account's firewalls with their inbound rules and the machines they protect, plus the quota. Scope: vms.read.

• get_firewall

  One firewall: its rules and the machines it protects, with the status of each (applied, pending, error). Scope: vms.read.

• create_firewall

  Creates a reusable set of inbound and outbound rules. Without rules it starts with SSH (TCP 22) and ping open from anywhere; pass rules: [] for none. Nothing is filtered until it is attached to a machine. Scope: network.write.

• update_firewall

  rules REPLACES the whole list and is applied at once to every attached machine: send every rule to keep, not only the new one (read get_firewall first). Removing the SSH rule can lock the agent out. pending_machines counts machines not…

• delete_firewall

  Deletes a firewall attached to no machine. Refused while it protects one: detach it first. Scope: network.write.

• attach_firewall

  From now on the machine refuses inbound traffic no rule of its firewalls opens; enforced within about ten seconds of the call. Up to 5 firewalls per machine, rules add up. Refused for a machine with no public address. Attaching again is…

• detach_firewall

  Removes the firewall's rules from the machine; with no other firewall it accepts all inbound traffic again. Scope: network.write.

• list_load_balancers

  The account's load balancers with their addresses, pools, targets and health, listeners and rules, certificates; plus the quota. Scope: vms.read.

• get_load_balancer

  One load balancer in full: status, ipv4/ipv6 to put in DNS, pools with targets (health, sessions, response time), listeners with rules in evaluation order, certificates. Scope: vms.read.

• get_load_balancer_metrics

  Requests by response class, connections, bytes in/out and peak sessions over 1h (1-minute buckets), 24h (15 minutes) or 7d (1 hour), plus totals. Scope: vms.read.

• create_load_balancer

  Orders a load balancer in a private network, billed by the hour from now (the account needs 24 hours of credit), or by the month (7.30 CAD) or the year (73 CAD) with billing. A monthly or yearly period comes out of credit if it covers it…

• delete_load_balancer

  Deletes the load balancer, its address and its configuration; billing stops. Target machines are not touched. Ask the user first. Scope: vms.destroy.

• upsert_pool

  Without pool: creates one (name required, up to 10). With pool: changes only the fields given. Deleting a pool also removes its targets and the rules leading to it. Scope: network.write.

• delete_pool

  Removes a pool, its targets and the rules leading to it. Scope: network.write.

• set_targets

  REPLACES the pool's targets with exactly this list (up to 25): send every target to keep. Each machine must be in the load balancer's private network (connect_vpc first). drain: true stops new connections to a target before maintenance…

• upsert_listener

  An entry port on the load balancer's address. Without listener: opens one (port and protocol required, up to 5; 22 and 8402 are reserved). https gets certificates for its rules' hostnames; redirect_https makes port 80 redirect to it. A…

• delete_listener

  Closes the port and removes its rules. Scope: network.write.

• set_rules

  REPLACES the http/https listener's rules: each sends a hostname and/or path prefix to an http pool (up to 25). They are evaluated most specific first, whatever the order sent; anything unmatched goes to the default pool. No wildcard…

• list_databases

  The account's managed databases with their status, private and public addresses, disk, backups and billing (no passwords), plus the quota. Scope: vms.read.

• get_database_options

  What can be ordered: engines and versions, extensions, sizes with their hourly, monthly and yearly prices per region, and the price of disk beyond what a size includes. Call it before create_database and show the price to the user. Scope…

• get_database

  One database: status, connection (host, private ip, port, defaultdb and ffxfadmin; on Valkey no database and default), public access and sources, disk usage and cap, users with their databases (null: all) and read_only, logical databases…

• get_database_credentials

  Every user with its password, databases and read_only, and its own connection_uri; ready-made connection strings for the admin (private network, and public when open). Treat the result as a secret: give it to the application through an…

• get_database_metrics

  CPU, memory, connections and commits per minute (commands per minute on Valkey) over 1h (1-minute buckets), 24h (10 minutes) or 7d (1 hour); only buckets with samples. Scope: vms.read.

• get_slow_queries

  The 20 queries with the most total time (pg_stat_statements, values replaced by $1, $2…; performance_schema on MySQL and MariaDB, values replaced by ?): calls, total and mean time, rows. A starting point for indexes. On Valkey, the…

• list_database_backups

  Completed backups (weekly full, daily differential; the transaction log is archived continuously) and the restore window: any minute in it can be restored with restore_database. On Valkey, snapshots (rdb) every hour when there were…

• create_database

  Orders a managed database (PostgreSQL, MySQL, MariaDB or Valkey) in a private network, billed by the hour from now (the account needs 24 hours of credit), or by the month (730 hours) or the year (ten months) with billing…

• restore_database

  Creates a NEW database, billed like any other, holding the data as it was at a given minute of the source's restore window (list_database_backups; Valkey: the last snapshot before that minute), with the same users and passwords. The…

• fork_database

  Creates a NEW database, billed like any other, whose data is an instant copy of this one (for a test or staging copy), with the same users and passwords. It stays linked to its source for up to 7 days (no backups of its own, the source…

• detach_database_fork

  Gives a linked fork its own copy of the data; the database is unavailable for about 30 seconds, then its backups start. Scope: vms.action.

• resize_database

  Moves the database to a LARGER size (more vCPU, memory and included disk; see get_database_options): same data, users, names and addresses. It restarts: one to two minutes of downtime, status maintenance meanwhile (resize.to_gb set), then…

• delete_database

  DELETES the database and its data; billing stops. Ask the user first. confirm_name must be the database's exact name. Its backups stay 7 days (restore_database can bring it back meanwhile). Refused while a linked fork depends on it…

• update_database

  Any of: the name; the weekly maintenance window (day 0 = Sunday, start hour, Montréal time, 2 hours long; minor updates, usually under 2 minutes of downtime); the disk cap (the disk grows by itself at 80 % full, up to the cap; disk beyond…

• set_database_public_access

  Opens the database to the internet at its public host and port (IPv4 and IPv6, TLS and password still required), or closes it. sources restricts who may connect (IPv4/IPv6 addresses or ranges); omitted it is left as is, null or [] means…

• backup_database

  Starts a differential backup within a minute, e.g. before a risky migration. Not available on a linked fork. Scope: backups.write.

• create_database_user

  Creates a user with a generated password, returned here once (also readable with get_database_credentials); live within seconds. Up to 30 users. It connects only to its databases (every logical database, future ones included, when…

• set_database_user_access

  Sets the logical databases a user connects to (databases: a list, or null for all of them, future ones included) and whether it only reads; a field left out keeps its value. Applied within seconds; where its access is removed or lowered…

As primeiras 60 de 71, na ordem do servidor.