Tools
Everything an AI assistant can do through Pyo's MCP server.
Pyo's MCP server has 132 tools. They act as the person who approved the assistant, with the same permission checks as the dashboard.
An assistant only sees the tools its access allows: 30 with Read only, 39 with Deploy and all 132 with Full. See access.
Most tools do the same as an API operation, and take the same fields. Its reference page lists them.
Common tools
Start with these. They cover what most conversations need.
| Tool | What it does | Inputs | Access | API |
|---|---|---|---|---|
list_teams | Teams the user belongs to. | None | Read | Reference |
list_services | Services of a team: status, replicas, domains, live deployment. | team | Read | Reference |
get_service | One service in detail: source, settings, domains, resources, runtime, recent deployments. | service_id (required) | Read | Reference |
list_deployments | A service's deployments, newest first. | service_id (required), limit | Read | Reference |
get_deployment | A deployment's status, why it failed, and its build and rollout logs. | deployment_id (required) | Read | Reference |
get_service_logs | Recent output of the service's running replicas. | service_id (required) | Read | Reference |
list_servers | Clusters and servers of a team, with CPU and memory. | team | Read | Reference |
deploy_service | Deploy the service with its current settings (Git services build their branch head), or retry a failed deployment. Returns the deployment id. | service_id (required), retry_of | Deploy | Reference |
rollback_service | Go back to an earlier successful deployment (the previous one when omitted). | service_id (required), to_deployment_id | Deploy | Reference |
scale_service | Set how many replicas run (1 or more), without redeploying. Replicas that don't fit on the servers wait until there's room, and the result has a warning saying so. A very large change (20 or more replicas, and over 10 times the current count or twice what fits) is refused until confirmed: ask the user first, then call again with confirm: true. | service_id (required), replicas (required), confirm | Deploy | Reference |
restart_service | Restart the service's replicas one at a time. | service_id (required) | Deploy | Reference |
create_project | A new, empty project (a group of services). Returns its id. | team, name (required), description, namespace | Full | Reference |
create_service | Create a service from a container image and deploy it. The project is found by name or id, or created when it doesn't exist. Returns the service id, deployment id and URL. 20 or more replicas is refused until confirmed: ask the user first, then call again with confirm: true. | team, project, name (required), image (required), port, replicas, confirm, env, free_domain, domain, region, server_tags | Full | Reference |
add_domain | Serve a service on a hostname, or on a free nip.io domain. It goes live in place, without a redeploy (a service never deployed gets it with its first deploy). | service_id (required), hostname, free_domain | Full | Reference |
stop_service | Stop the service (scale to zero). Settings and data are kept. | service_id (required) | Deploy | Reference |
Account
| Tool | What it does | Access | API |
|---|---|---|---|
get_me | The account these credentials act as: an API key's owner, or the person who connected an assistant. | Read | Reference |
get_meta | Its version, and the features a hosted Pyo turned on (none when self-hosted). No API key needed. | Read | Reference |
State
| Tool | What it does | Access | API |
|---|---|---|---|
get_state | One snapshot of every team you're in (or the API key's teams): their projects, services and deployments, clusters and servers, domains, certificates, volumes, backups, registry logins, builders and recent activity. The dashboard renders from this, and scripts can read it too. | Read | Reference |
Observability
| Tool | What it does | Access | API |
|---|---|---|---|
get_access_logs | The web server's log of requests to a team's websites, newest first, filtered like the traffic figures. | Read | Reference |
get_metrics | CPU, memory, disk and network over time for a team's servers and services, sampled once a minute. Servers also get load and disk reads and writes. Each series lines up with times, with null where nothing was measured. | Read | Reference |
get_observability | Visits, requests, errors, response times, countries, pages and referrers for a team's websites: all of them, one project's, one service's, or one domain. | Read | Reference |
get_observability_settings | Whether observability is on for this installation (it's off on a new install on a server with under 3.5 GB of memory), and what it keeps of visitors' IP addresses. small: the server Pyo runs on has under 3.5 GB of memory now. The installation's admin turns it on or off with setObservability. | Read | Reference |
get_service_connections | Connections to a service's public TCP and UDP ports through the gateway: per port, over time, by country and outcome ("ok", "blocked" by an IP rule, "limited" by the connection limit, "unreachable" when the service didn't answer), and the latest 50. A connection is counted when it closes (a UDP exchange after a minute without packets). Ports that bypass the gateway aren't counted. | Read | Reference |
set_observability | Turn observability (analytics, request logs, firewall and WAF history, usage history) on or off for the whole installation. Only the installation's admin can do this. Off stops storing and showing history and frees its memory. What was stored is kept for when it's back on. Live logs, current usage, deploys and the firewall keep working. | Full | Reference |
Services
| Tool | What it does | Access | API |
|---|---|---|---|
create_service_from_spec | Create a service and deploy it. | Full | Reference |
delete_service | Delete a service and stop it everywhere. Its volumes are kept, detached, unless deleteVolumes. | Full | Reference |
get_service_preview | Whether a service's site can be shown in the dashboard's preview frame: it answers, and doesn't forbid framing (X-Frame-Options, CSP frame-ancestors). Only the service's own first domain is asked, and never a private address. | Read | Reference |
get_service_traffic | Its domains and the addresses they point at, the servers that take public traffic and their regions, where its replicas run, and anything that keeps visitors from reaching it. | Read | Reference |
reveal_secret | Show a secret environment variable's value. Admins only, and logged. Returns a secret. | Full | Reference |
save_service | Save a service's configuration. With deploymentId, roll it out too. Changes to its domains go live without a deploy once it has been deployed. | Full | Reference |
set_autoscale | Turn autoscaling on or change it. Null turns it off. | Deploy | Reference |
Deployments
| Tool | What it does | Access | API |
|---|---|---|---|
cancel_deployment | Cancel a deployment that hasn't finished. | Deploy | Reference |
create_deploy_hook | Create a deploy hook: a secret URL that deploys a service, or every service of a project. | Full | Reference |
remove_deploy_hook | Remove a deploy hook. Its URL stops working. | Full | Reference |
reveal_deploy_hook | Show a deploy hook's URL path (its secret token). Developers and up. Returns a secret. | Full | Reference |
Cron jobs
| Tool | What it does | Access | API |
|---|---|---|---|
list_cron_runs | A cron job's recent runs, newest first: when each started and finished, how it ended, and the end of what it printed. | Read | Reference |
remove_cron_job | Remove a cron job from a service. | Full | Reference |
run_cron_job | Run a cron job now. | Deploy | Reference |
save_cron_job | Add a cron job to a service, or change one (by id). | Full | Reference |
Databases
| Tool | What it does | Access | API |
|---|---|---|---|
add_database_backup | Checks the admin login from the cluster, makes a backup user with it, then saves the database and its schedule. Team owners and admins. | Deploy | Reference |
check_database_import | Everything startDatabaseImport checks, without importing. The cluster connects to the source, which must be reachable from there, and checks the login, TLS, its version, its extensions and that every table is readable. It also checks that the database to import into is empty. Owners and admins. | Full | Reference |
connect_database | Connect a service to a database on its cluster. Postgres adds DATABASE_URL (through the pooler when it's on) and DATABASE_URL_DIRECT, plus DATABASE_URL_READ with read replicas. MySQL adds DATABASE_URL (mysql://, to the primary), MYSQL_HOST, MYSQL_PORT, MYSQL_USER, MYSQL_PASSWORD and MYSQL_DATABASE. KV adds REDIS_URL (which follows failover and sends reads to read replicas), plus KV_REST_API_URL, KV_REST_API_TOKEN and KV_REST_API_READ_ONLY_TOKEN (reads only) with its REST API. With env, it adds only those, under the service's own names. With prefix, it adds all of them with that prefix, so two databases' variables don't clash. They're secret variables and apply on the service's next deploy. | Full | Reference |
create_database | Create a Postgres, MySQL or KV (Redis-compatible key-value, Valkey) database on one of the team's clusters. The first database on a cluster also sets up databases there, which takes a minute or two. size sets CPU and memory per instance: "small" (0.25 CPU, 512 MiB, 5 GiB), "medium" (0.5, 1 GiB, 10 GiB), "large" (1, 2 GiB, 20 GiB) or "xlarge" (2, 4 GiB, 50 GiB). MySQL needs 1 GiB, so "medium" or larger. instances is 1 for a single server with no automatic failover. For high availability, use a primary and 2 to 8 replicas for Postgres (instances 3 to 9), 2 to 4 for KV (3 to 5), or 2 or 4 for MySQL (3 or 5). A replica takes over when the primary's server is lost. MySQL has no pooler, read replicas or REST API. Developers and up. | Full | Reference |
delete_database | Delete a database and its data. Refused while deletion protection is on or a service is connected. Owners and admins. | Full | Reference |
delete_database_backup | Stop backing up a database. Its existing backups stay in storage. | Full | Reference |
disconnect_database | Disconnect a service from a database: removes the variables it added (on the next deploy). | Full | Reference |
get_database | One Postgres, MySQL or KV database, as listDatabases lists them. | Read | Reference |
get_database_analytics | Its health over time, sampled about every 30 seconds. Per member (members: each instance, read replica, pooler and KV endpoint), its role, server, and series for cpu in cores, memoryMi, diskUsedGi (only where the volume reports its own space), connections and lagSeconds. For the whole database, transactions (Postgres), queries (MySQL), poolerQueries and commands (KV) per second, connections (primary, replicas, limit, and the pooler's poolerActive, poolerWaiting, poolerServersActive, poolerServersIdle and poolerMaxWaitSeconds), cacheHitRatio, keyHitRatio, evictions per second, routing (KV reads per second served by replicas or the primary, and fallbacks to the primary) and storage (sizeBytes, memoryUsedBytes, maxmemoryBytes, keys). Each series lines up with times, with null where nothing was measured. A series is null when the database doesn't report it. | Read | Reference |
list_database_extensions | The extensions Pyo's databases offer: Postgres extensions and KV modules. Each one's name, engine ("postgres" or "valkey"), section ("managed" ones are turned on with setDatabaseExtensions, and the database's owner adds "additional" ones with CREATE EXTENSION), a one-line description, its docs, its edition when it's a limited build, hidden when it's obsolete (shown only while a database has it on), and whether turning it on or off restarts the database (every KV module does). | Read | Reference |
list_databases | A team's Postgres and MySQL databases and Valkey (Redis-compatible) KV databases: engine, version, size, instances, the cluster's status (phase, primary, ready instances, failovers), the services connected to each and their variables, passwords masked. | Read | Reference |
restore_database | Restore a Postgres or MySQL database into a new one, on the same cluster with the same size and availability: at targetTime (RFC 3339, within what its backups cover: from status.backup.firstRecoverabilityPoint to now for Postgres, to status.backup.lastSuccessfulBackup for MySQL) or the latest when omitted. Developers and up. | Full | Reference |
retry_database_import | Runs a failed import again, from the same source. After a failure, Pyo keeps the URL's password (sealed), and an uploaded dump for PYO_DATABASE_IMPORT_RETENTION_HOURS (24 by default). A new database made for the import is made again. A database that already existed must still be empty. Owners and admins. | Full | Reference |
reveal_database_password | Show a database's password, and a KV database's REST token and read-only token. Owners and admins. Recorded in the activity log. | Full | Reference |
run_database_backup | Back up a database now. | Full | Reference |
set_database_backups | Turn a database's backups on or off, to the team's Backups destination. Postgres: every change archived continuously and a base backup daily, so it can be restored to any moment in the last retentionDays. MySQL: a full backup daily with the changes since the last one, so it can be restored to any moment in the last retentionDays up to its latest backup. Both are stored in the team's folder of the destination. KV: a snapshot scheduled daily or weekly, the last keep kept, listed in Backups with the team's other databases. Developers and up. | Full | Reference |
set_database_extensions | Turn a Postgres database's extensions or a KV database's modules on and off. extensions is the full list to have on, from listDatabaseExtensions with section "managed" for its engine. The database's owner adds Postgres's "additional" ones with CREATE EXTENSION. Some load a library, which restarts the database (restart). With high availability, it restarts one instance at a time and the primary hands over last. Every KV module restarts the database and needs KV 8.1. Turning a Postgres extension off drops it (DROP EXTENSION, without CASCADE). Postgres refuses while other objects depend on it, and the database's extensions shows this. Developers and up. | Full | Reference |
start_database_import | Copies a Postgres or MySQL database into one of the team's: a new one (newDatabase), or an empty one (databaseId). From url (the source must be reachable from the cluster) or uploadId (a dump uploaded first). It's checked first, as checkDatabaseImport. Then the database's import shows how it goes ("pending", "running", "succeeded" or "failed", with a phase), and its logs have the import's output. Writes made to the source during the import aren't copied. Owners and admins. | Full | Reference |
update_database | Change a database's size, instances, storage, version, pooler, REST API, read replicas or deletion protection. Instances are 1 for a single server, or for high availability 2 to 8 replicas for Postgres, 2 to 4 for KV, or 2 or 4 for MySQL. MySQL instances only grow. Storage only grows and the version only goes up. Only owners and admins can turn deletion protection off. Connected services get changed variables on their next deploy. | Full | Reference |
update_database_backup | Change a backed-up database's schedule, or pause its backups. | Full | Reference |
Backups
| Tool | What it does | Access | API |
|---|---|---|---|
remove_team_backup_settings | Stop a team's backups. | Full | Reference |
reveal_backup_password | Show the team's backup repository password, to restore without Pyo. Logged. Returns a secret. | Full | Reference |
run_team_backup | Back up now: one service or volume (by id), or everything selected. | Full | Reference |
save_team_backup_settings | Back up a team's volumes and databases: where, how often and how many to keep. A blank secret keeps the saved one. | Full | Reference |
set_team_backup_selection | Choose what a team backs up: "all" but exceptions, or only the "selected" ones. | Full | Reference |
Domains
| Tool | What it does | Access | API |
|---|---|---|---|
remove_certificate | Remove an uploaded certificate no domain uses. | Full | Reference |
replace_certificate | Replace a certificate with a renewed one. Services using it switch over in place, without a redeploy. | Full | Reference |
upload_certificate | Upload a certificate for the team's domains: the chain and its private key, PEM. | Full | Reference |
verify_domain | Check a domain's DNS now. | Full | Reference |
Volumes
| Tool | What it does | Access | API |
|---|---|---|---|
add_volume | Add a persistent volume to a service, mounted at mountPath from its next deploy. | Full | Reference |
attach_volume | Attach a detached volume to a service on its cluster, mounted at mountPath from the service's next deploy. A service of another project (or any project, for a volume whose project was deleted) gets the volume moved into its project first, data, id and backups kept: the volume shows moving until its cluster has moved it, then it's attached (or moveError says why not). A volume can't move to another cluster. A volume on a one-server storage class keeps its data on that server, so the service must be able to run there. | Full | Reference |
delete_volume | Delete a detached volume and its data, for good. Admins only. The volume stays listed, being deleted, until its cluster confirms. Also deletes a volume Pyo has no record of, listed in its cluster's strayVolumes: pass that volumeId as id, with clusterId. | Full | Reference |
delete_volume_snapshot | Deletes one of a volume's backups from the team's restic repository, with a short job on its cluster that waits for the volume's other backups and restores. The space it held is freed shortly after. Deleting one that's already gone succeeds. Owners and admins. | Full | Reference |
detach_volume | Detach a volume from its service: unmounted from the next deploy. The volume and its data stay, to attach again or delete. | Full | Reference |
list_volume_snapshots | Its backups in the team's restic repository, newest first, read from its cluster on request (a short job there). Takes up to two minutes. | Read | Reference |
resize_volume | Grow a volume to sizeGi GB (never shrink it). Owners and admins. Only on a storage class that can grow volumes (allowVolumeExpansion). A local-path volume has no size limit of its own, so there's nothing to grow. The volume shows resizeTo until its cluster reports the new size. Some storage classes finish when its service next starts. | Full | Reference |
restore_volume | Restore a volume from one of its snapshots (see listVolumeSnapshots), replacing its data. Owners and admins. Its service stops while the restore runs and starts again afterwards, even when it fails. The snapshot is restored next to the data and swapped in only once complete, so a failed restore leaves the data as it was. The volume needs room for the snapshot while this runs. The volume shows restore while it runs, then restoredAt, or restoreError with why it failed. | Full | Reference |
Teams
| Tool | What it does | Access | API |
|---|---|---|---|
create_team | Create a team. You become its owner. | Full | Reference |
delete_team | Delete a team and everything in it. Owners only. | Full | Reference |
invite_member | Invite someone to a team by email, as owner, admin, developer or viewer. | Full | Reference |
remove_member | Remove a member from a team, or revoke an invite. | Full | Reference |
resend_invite | Make a new invite link (the earlier one stops working) and mail it, or with send: false only return it to share. | Full | Reference |
rotate_data_key | Give the team a new data key and seal its stored secrets again with it. | Full | Reference |
update_member_role | Change a member's role: "owner", "admin", "developer" or "viewer". | Full | Reference |
update_team | Rename a team, change its description, default cluster or preferences. | Full | Reference |
Projects
| Tool | What it does | Access | API |
|---|---|---|---|
delete_project | Delete a project and its services. Its volumes are kept, detached, unless deleteVolumes. | Full | Reference |
import_compose | Create a project and its services from a Docker Compose plan (from POST /compose/preview, edited if you like), and deploy them in order. | Full | Reference |
preview_compose | What importing a Compose file would create: a project, and a service per Compose service with its source, ports, variables, volumes and health check, in deploy order. It adds a warning for each thing that isn't carried over. Nothing is created. Edit the plan if you like, then pass it to the importCompose operation. | Read | Reference |
set_project_isolation | Isolate a project's network: only its own services and the web server reach them. | Full | Reference |
update_project | Rename a project or change its description. | Full | Reference |
Clusters
| Tool | What it does | Access | API |
|---|---|---|---|
create_cluster | Add a cluster. The response has the command that connects its first server. | Full | Reference |
discard_cluster_draft | Discard a cluster that was added but never connected. | Full | Reference |
remove_cluster | Remove a cluster from Pyo. Its servers keep running. | Full | Reference |
rotate_agent_token | A new connect command for a cluster. The old one stops working. | Full | Reference |
set_cluster_free_domain_address | Where the cluster's free domains (<name>.<ip>.nip.io) point: "public" (its public IP, for everyone) or "tailscale" (its Tailscale IP, only from the tailnet). With "tailscale", services are reachable from your tailnet on ports 80 and 443 whether or not the cluster takes public traffic. Free domains on it move to the new address in place. By default it's public when the cluster takes public traffic and a server has its public IP itself (on a network card or as its ExternalIP) or a load balancer gives it one. Behind NAT, it's Tailscale. | Full | Reference |
set_cluster_public_traffic | Whether a cluster takes public traffic: the gateway answers on every server but build-only ones (ports 80 and 443, and public TCP and UDP ports through it). Off: services are only reachable inside the cluster. | Full | Reference |
set_cluster_web_proxy_protocol | Read the visitor's address from a PROXY protocol header on the cluster's websites (ports 80 and 443), for a cluster behind another load balancer. Connections without one still work. | Full | Reference |
update_cluster | Rename a cluster or change its description, location or tags. | Full | Reference |
Servers
| Tool | What it does | Access | API |
|---|---|---|---|
cancel_server_removal | Stop a server removal that's waiting or failed. The server takes work again. | Full | Reference |
join_server | A script that joins a new server to a cluster (its token is valid for an hour). | Full | Reference |
remove_server | Take a server out of its cluster, moving its services off first. force removes it at once, for a machine that's gone. | Full | Reference |
set_server_maintenance | Put a server in maintenance (its services move elsewhere), or bring it back. | Full | Reference |
set_server_region | Set a server's region: a name ("US East", "Singapore, SG"), or "" to detect it again. | Full | Reference |
set_server_regular_services | On a GPU server: also run regular services, or keep it for GPU work. Null goes back to the default. | Full | Reference |
set_server_schedulable | Whether new work may be placed on a server. | Full | Reference |
set_server_tags | Set a server's tags. | Full | Reference |
Firewall
| Tool | What it does | Access | API |
|---|---|---|---|
add_firewall_rule | Add an access rule to the end of a service's firewall. Allow and block rules are checked in order, and the first match decides. With no match, the client gets in. Rate limit rules then apply to websites, and each matching one counts the request, per client address and per server. | Full | Reference |
add_waf_rule | Add a custom web application firewall rule to a service. Custom rules run in order before the OWASP rules, while the service's WAF is on. It goes last in the list. | Full | Reference |
get_firewall_blocked | What a service's firewall turned away: website requests (a 403, or a 429 over a rate limit) and connections to TCP ports through the gateway. It shows them over time (series), per rule, per source, country and website path, and the latest 50. A rule is its id, or "blocked" or "rate-limit" when no rule matches the source now. UDP ports and ports that bypass the gateway aren't counted. | Read | Reference |
get_waf_hits | What a service's web application firewall found: requests it blocked, and ones it only logged in log-only mode. It shows them over time (series), per rule, domain, source, country and path, and the latest 50 with the rules that matched. custom shows how often each custom rule matched, by its number. Query strings are kept with sensitive values redacted. | Read | Reference |
remove_firewall_rule | Remove a firewall access rule. | Full | Reference |
remove_waf_rule | Remove a custom web application firewall rule. | Full | Reference |
reorder_firewall_rules | Put a service's firewall access rules in this order: allow and block rules are checked from the first. List every one of its rules. | Full | Reference |
reorder_waf_rules | Put a service's custom web application firewall rules in this order. List every one of its rules. | Full | Reference |
set_service_waf | A service's web application firewall (WAF) settings, for all its domains. "log" or "block" turns it on: the WAF starts on the service's cluster with the first service there that turns it on, and stops after the last one turns it off. | Full | Reference |
update_firewall_rule | Change a firewall access rule. Omitted fields stay as they are (as in addFirewallRule). | Full | Reference |
update_waf_rule | Change a custom web application firewall rule. Omitted fields stay as they are. | Full | Reference |
Builders
| Tool | What it does | Access | API |
|---|---|---|---|
create_builder | Add a builder that builds images from Git on one of the team's servers (kind "cluster"). | Full | Reference |
remove_builder | Remove a builder. | Full | Reference |
rotate_builder_token | A new token for a builder. The old one stops working. | Full | Reference |
Images
| Tool | What it does | Access | API |
|---|---|---|---|
create_registry_token | Create a token to push (or only pull) the team's own images in Pyo registry, as <host>:5443/<team>/<project>/<service>:<tag>. These are slugs, and <host> is the installation's tailnet name. A push needs that project and service to exist. It works over HTTPS from a device on the installation's tailnet. Team owners and admins. The token comes back only once, with the username, address and team path to log in and push with. | Full | Reference |
get_registry | Pyo's registry, where the images it builds from Git and the team's own pushes are kept, as one team sees it: its address, where it runs (when the team has that cluster), whether Pyo reaches it, the space the team's images take, the team's repositories (tags, size, last build) and its cleanup: how many deployments' images are kept per service, and when images were last removed (within the last week). Read only. | Read | Reference |
list_images | Every image a team's services deployed, by repository, with each version's deployments, the services that used it and whether one runs it now. | Read | Reference |
remove_registry_login | Remove a registry login. | Full | Reference |
revoke_registry_token | Revoke a Pyo registry token. Team owners and admins. | Full | Reference |
save_registry_login | Save a login for a private registry, one per host. Pyo checks it with the registry first. | Full | Reference |
GitHub
| Tool | What it does | Access | API |
|---|---|---|---|
get_github_build_file | API key access: read. | Read | Reference |
github_install_url | The GitHub page where a team installs the GitHub App. | Full | Reference |
list_github_repos | The repositories the team's GitHub installations can see, to deploy from. | Read | Reference |
remove_github_installation | Unlink a GitHub installation from a team. | Full | Reference |
Integrations
| Tool | What it does | Access | API |
|---|---|---|---|
check_secret_import | The key names at a path of a secret store, to check it before importing. Never values. | Full | Reference |
remove_integration | Remove a team integration's key. | Full | Reference |
save_integration | Save a team integration's key (kind "tailscale": an auth key, used when servers join over Tailscale). | Full | Reference |