Let AI assistants manage your SailOps infrastructure through natural language.
The SailOps MCP Server implements the Model Context Protocol (MCP), a standard that allows AI assistants such as Claude, Cursor, and other MCP-compatible clients to interact with external systems using structured tools.
With the SailOps MCP Server, you can ask your AI assistant to deploy a container, create a database, expose a service through a gateway, or apply an entire infrastructure stack — all without writing a single API call yourself.
X-Api-Key header, or Authorization: Bearer <key> (same key used for the Public API)
Tools available
Every Public API endpoint except the health check, plus the console's project operations: metrics, logs, events, policies, alerts, audit
The MCP server is scoped to a project, just like API keys. Each key provides access only to the project it was created in. A key restricted to some scopes only lists and runs the tools those scopes allow: a key holding containers:list alone can read containers but not change them.
Replace <your-api-key> with an API key created in Project Settings > API Keys.
Use environment variables
To avoid committing your API key, reference an environment variable in your shell profile instead of hardcoding the key value. You can also add .claude/mcp.json to your .gitignore.
Any MCP-compatible client that supports Streamable HTTP transport can connect to the SailOps MCP server. The configuration format varies by client, but the connection details are always the same:
URL: https://app.sailops.io/api/mcp
Transport: Streamable HTTP (not SSE)
Header: X-Api-Key: <your-key>, or Authorization: Bearer <your-key> for clients that only send a bearer token
Once connected, you can manage your infrastructure through natural language:
You: Deploy a Node.js API with a PostgreSQL database and expose it on api.example.com
Claude: I'll set that up for you. Let me create the resources step by step.
Creating the PostgreSQL database main-db (version 16, standard class, 20Gi storage)...
Creating the container api from node:20-alpine with 500m CPU and 512Mi RAM...
Connecting api to main-db — this will inject MYDB_PGHOST, MYDB_PGPORT, MYDB_PGUSER, MYDB_PGPASSWORD, and MYDB_PGDATABASE environment variables...
Creating gateway main-lb (class: gw-small)...
Adding domain api.example.com with HTTPS listener...
Creating HTTP route pointing all traffic to port 3000 on api...
Everything is ready. Your API will be available at https://api.example.com once the TLS certificate is provisioned (usually within a few minutes). Push your Node.js image to the container registry and update the container image to start serving traffic.
All tools follow the naming convention {domain}_{action}. Most tools correspond directly to a Public API endpoint; the others expose the project operations of the console (metrics, logs, events, policies, alerts, audit log).
Get the whole project in one call: every workload, managed service, gateway, route, website and shared volume with its status, plus the connections between them
Get a website's replica settings: fixed count or autoscaling bounds, current and desired replicas
websites_autoscaling_update
Set a website's replica settings: a fixed replica count, or autoscaling bounds
websites_set_domain
Attach a custom domain to a website through a gateway; the route is created or updated for you
websites_remove_domain
Detach the custom domain from a website and remove its gateway route
websites_republish
Regenerate a website's serving configuration and roll it out again from the current files
websites_deployments_list
List a website's deployments (versions) with status, file count, size and activation time
websites_list_files
List files and directories in a website's storage
websites_read_file
Read the text content of a file from a website's storage
websites_write_file
Write text content to a file in a website's storage (creates or overwrites it)
websites_create_directory
Create a directory in a website's storage
websites_delete_file
Delete a file or directory from a website's storage
websites_get_storage_stats
Get storage usage statistics for a website (total size and file count)
websites_logs
Get a website's web server logs (supports lines, since, query and stream filters)
websites_metrics
Get a website's resource metrics (CPU, memory, network) as a time series (default: the last hour)
websites_route_metrics
Get the HTTP traffic metrics (request rate, success and error rates, upstream response time) of a website: its Quick Expose and custom-domain routes summed across gateways. Defaults to the last hour.
Replace a gateway's definition: class, listeners and proxy protocol (listeners absent from the body are removed)
gateways_update_class
Change a gateway's class (size)
gateways_delete
Delete a gateway and all its routes
gateways_classes_list
List the gateway classes (sizes) with their capacity and limits
gateways_default_get
Get the project's default gateway and the wildcard domain under which resources get an automatic hostname
gateways_metrics
Get traffic metrics of every route of a gateway as fixed intervals of bucketSeconds: counts are per interval, requestsPerSecond is requests.total / bucketSeconds, successRate and errorRate are null without traffic, latency is upstream response time in ms (null when no request completed, a lower bound when latencyPxxOverflow is true). TCP and TLS routes report tcpConnectionsTotal. Defaults to the last hour. gatewayRequests independently reports listener requests, responses and 4xx/5xx, including gateways without routes; gatewayAnswered estimates gateway-generated responses only when routed and listener counters are observed.
Validate a route definition without creating it: validation errors and conflicts with existing routes
routes_backends_list
List the services a route can send traffic to (containers, websites, managed services) with their ports and health
routes_metrics
Get traffic metrics of one route as fixed intervals of bucketSeconds: counts are per interval, requestsPerSecond is requests.total / bucketSeconds, successRate and errorRate are null without traffic, latency is upstream response time in ms (null when no request completed, a lower bound when latencyPxxOverflow is true). TCP and TLS routes report tcpConnectionsTotal. Defaults to the last hour. Optional kind: HTTPRoute, GRPCRoute, TCPRoute or TLSRoute.
routes_policy_attach
Attach an existing security, client-traffic or backend-traffic policy to a route
routes_policy_detach
Detach a policy from a route (the policy itself is kept)
Update flavor, compute class, storage, or I/O timing config
services_postgresql_delete
Delete a PostgreSQL service
services_postgresql_configuration_update
Update PostgreSQL parameters (max_connections required); omitted I/O timing settings are preserved, other omitted settings return to defaults. I/O timing changes apply without a restart
services_postgresql_restore
Restore a backup or a point in time into a new service; the source is left untouched
services_postgresql_roles_list
List the managed roles of a PostgreSQL service with their access level per database
services_postgresql_roles_create
Create a PostgreSQL role with an access level per database (password returned once)
services_postgresql_roles_update
Set the access level per database of a PostgreSQL role
services_postgresql_roles_delete
Delete a PostgreSQL role
services_postgresql_schemas_list
List the schemas of a PostgreSQL service with their owner
services_postgresql_schemas_create
Create a schema in a PostgreSQL service, optionally owned by one of its managed roles
services_postgresql_schemas_delete
Drop a schema of a PostgreSQL service and every table it holds
services_postgresql_logs
Get the server logs of a PostgreSQL service (timestamp, level, instance and message per line)
services_postgresql_metrics
Get the engine metrics of a PostgreSQL service over a window: connections, transactions, I/O, replication, storage and WAL
Search workload, metric and Trivy incidents in the project, ordered by severity then newest start time. Returns bounded evidence and the matching total; resolved incidents require includeResolved=true.
This tool requires alert-center/incidents:list. Keys restricted to the historical alerts:list scope need an explicit scope update. API keys have no personal read state, so status=read and status=unread are unavailable. Results include items, total, limit and truncation metadata; they do not include project-wide statistics.
List the project's audit log, newest first: actor, operation, target resource, time and origin. A background change ends with an entry whose action is suffixed .applied or .failed. Each entry of a background change carries an operationId that ties the pair together (filters: from, to, action, operation, resource, limit)