MCP tools

The geass mcp command runs a Model Context Protocol (MCP) server that gives an AI agent the same operations as the command line. This page lists every tool the server can register. To run the server and connect an agent, read MCP server for AI agents.

What the server exposes

• Tools only. The server registers tools. It exposes no MCP resources and no prompts, and it advertises no completions. It does advertise the standard logging capability of the MCP Go SDK it is built on.
• Identity. The server’s name is geass, its title is Geass and its version is the command line’s version (dev when the build did not set one).
• Instructions. When an agent connects, the server sends it a short briefing (see Instructions).
• Transports. Standard input and output, or Streamable HTTP with --listen.
• 27 tools in all, 26 without delete_resource. Which of them an agent sees depends on its token (see Which tools an agent sees).

Every tool takes a JSON object. An argument that is not in the tool’s schema is rejected. A tool’s result is one block of text: indented JSON for every tool except get_logs, which returns plain lines. A result longer than 60,000 characters is cut and ends with ... (cut: the result was too long; narrow it with the tool's options).

Hints

Each tool carries the standard MCP annotations, so an agent host can decide when to ask you first. read-only means readOnlyHint: true. idempotent means idempotentHint: true. Every tool except delete_resource has destructiveHint: false, and every tool has openWorldHint: false.

Errors

When Geass refuses or fails a call, the tool returns an error result (isError: true) whose text is the message from the API, with a note in brackets added after a space, for three statuses:

Status	Text added
401	(the token is missing, wrong or expired)
403	(this token is not allowed to do that; do not retry)
404	(it does not exist, or is not visible to this token)

Any other status gives the API’s message alone, such as replicas must be a whole number from 0 to 1000. A deploy that fails while the agent waits for it is not an error: it is a result that says why (see deploy_app).

Overview

Tool	What it does	Needs	Hints
whoami	Says who the agent acts as and what it may do	Any token	read-only
list_projects	Lists projects	Any workspace token	read-only
list_apps	Lists services	Any token	read-only
get_app	One service’s settings and status	Any token	read-only
app_status	Whether a service is up, and what its last deploy did	Any token	read-only
app_deployments	A service’s deployment history	Any token	read-only
app_metrics	CPU, memory, network and requests over time	Any token	read-only
get_logs	Recent log lines of a service	Read only, with logs	read-only
deployment_logs	Build and deploy logs of one deployment	Read only, with logs	read-only
list_variables	Names of a service’s variables	Read only, with logs	read-only
list_databases	Lists databases	Any workspace token	read-only
get_database	One database’s settings and status	Any workspace token	read-only
database_connection	Host, port, database and user	Full access	read-only
list_buckets	Lists object storage buckets	Any workspace token	read-only
list_nodes	Lists the cluster’s nodes and their tags	Any workspace token	read-only
project_compliance	A project’s required node tags	Any workspace token	read-only
deploy_app	Deploys a service’s pending changes	Full access or project token
rollback_app	Returns a service to an earlier deployment	Full access or project token
scale_app	Sets the number of copies	Full access or project token	idempotent
set_sleep_when_idle	Makes a service sleep when idle, or stay running	Full access or project token	idempotent
set_variable	Sets a variable or secret	Full access or project token	idempotent
unset_variable	Removes a variable or secret	Full access or project token	idempotent
set_function_trigger	Adds or replaces a function’s schedule	Full access or project token	idempotent
remove_function_trigger	Removes a function’s schedule	Full access or project token	idempotent
invoke_function	Calls a function at its public address	Full access or project token
create_app	Creates a service from an image or a repository	Full access
delete_resource	Permanently deletes a resource	Full access, and --allow-delete	destructive

Needs uses the access names of the dashboard. Full access is the admin role, Read only, with logs is observer, and Read only is viewer. A project token is limited to one environment of one project. Where a row says Any token, a project token can use the tool for services in its own environment. Any workspace token means a project token cannot. Read only, with logs also covers Full access and a project token. See Users and API tokens.

Which tools an agent sees

A tool that a token cannot use is not registered at all, so the agent never sees it. The flags of geass mcp narrow this further. --read-only removes every tool that changes anything. It keeps database_connection for a Full access token, because that tool only reads. --allow-delete adds delete_resource.

Token	Tools
Read only (viewer)	whoami, list_projects, list_apps, get_app, app_status, app_deployments, app_metrics, list_databases, get_database, list_buckets, list_nodes, project_compliance (12)
Read only, with logs (observer)	The 12 above, plus get_logs, deployment_logs, list_variables (15)
Full access (admin)	All except delete_resource (26), and delete_resource with --allow-delete (27)
Full access with --read-only	The 15 of an observer, plus database_connection (16).
Project token	whoami, list_apps, get_app, app_status, app_deployments, app_metrics, get_logs, deployment_logs, list_variables, deploy_app, rollback_app, scale_app, set_sleep_when_idle, set_variable, unset_variable, set_function_trigger, remove_function_trigger, invoke_function (18)

A project token with --read-only gets the first nine tools in its list. Hiding a tool is a convenience. The API checks the token on every call, so a tool that is shown can still fail with a 403 for something the token may not reach, such as app_metrics with kind set to databases for a project token.

Instructions

On connecting, the server tells the agent what it is allowed to do. The text starts with Geass runs services, databases and buckets on a cluster. Use list_apps and app_status to see what is running. and then, depending on the token:

• For a project token: You act with a token limited to <project>/<environment>: you can only see and change services there.
• For a token that cannot change anything, or with --read-only: You can only read: nothing here can change anything.
• Otherwise: Changing a variable does not apply it: call deploy_app afterwards (or pass deploy=true). To deploy and know whether it worked, call deploy_app with wait=true: it returns the outcome, and says why if the build failed.

Every version ends with Secret values are never returned by any tool. Plain configuration is not secret and is returned as it is, so keep anything sensitive in a secret.

Services

whoami

Says who the agent acts as and what it may do.

• Arguments: none.
• Needs: any token. Calls: nothing. It returns what the server learned from GET /me when it started.
• Returns: the identity.

Field	Meaning
username	token:<token name>.
role	admin, observer or viewer. A project token reports admin.
scope	workspace or project.
project, environment	The project token’s limits. Only for a project token.
canMutate	Whether the token can change things.
canReadSensitive	Whether it can read logs and variable names.
authentication	token.

Example result

{
  "username": "token:CLI: geass CLI on laptop",
  "role": "observer",
  "scope": "workspace",
  "canMutate": false,
  "canReadSensitive": true,
  "authentication": "token"
}

list_projects

Lists the projects.

• Arguments: none.
• Needs: any workspace token. Calls: GET /projects.
• Returns: {"items": [...]}. Each item is a project summary: kind (project), name (the project’s ID, such as proj-1a2b3c4d), displayName, state, message, createdAt and details with environments, and region and requiredTags when set.

list_apps

Lists services with their state (ready, pending, failed, draft or deleting).

Argument	Type	Required	Meaning
project	string	no	Only services in this project (its ID).
environment	string	no	Only services in this environment, such as production.

• Needs: any token. A project token sees only its own environment. Calls: GET /apps with project and environment.
• Returns: {"items": [...]}. Each item has kind (app), name, displayName, project, environment, state, message, url, pendingChanges, createdAt and details. details holds port, source (image, git or function), the image, or the repository and branch, or the runtime, mode and triggers, replicas, and scaleToZero (idleSeconds, maxReplicas, active, message) when the service sleeps when idle.

Example call

{ "project": "proj-1a2b3c4d", "environment": "production" }

get_app

Shows one service’s settings and status.

Argument	Type	Required	Meaning
app	string	yes	The service’s name, as listed by list_apps.

• Needs: any token. Calls: GET /apps/{app}.
• Returns: {"resource": {...}, "spec": {...}, "status": {...}}. resource is the same summary as in list_apps. spec holds the service’s settings and status what the operator last reported.

Secret values are never included: spec.secretData lists their names with empty values. Plain configuration is not secret and appears in spec.configData and spec.env when the token is Full access. A Read only or Read only, with logs token gets them removed. The tool’s description says the same.

app_status

Shows whether a service is up and what its last deploy did. Call it to find out whether a deploy worked.

Argument	Type	Required	Meaning
app	string	yes	The service’s name.

• Needs: any token. Calls: GET /apps/{app}/status.
• Returns:

Field	Meaning
name, project, environment	Which service.
state	ready, pending, failed, draft or deleting.
message	Why the service is not ready, when it says.
url	Its public address, when it has one.
ready	true when state is ready.
pendingChanges	How many changes wait for a deploy.
replicas	desired, ready and available copies. Present once the service has a Deployment.
sleeping	true for a service that sleeps when idle and has stopped. It is waiting for a request, not down.
scaleToZero	For a service that asked to sleep when idle: idleSeconds, maxReplicas, and active and message once the operator has said whether it does and why not.
build	The latest build: name, phase (Pending, WaitingForCI, Running, Succeeded, Failed or Cancelled), failureReason, sourceRevision, ciState, startedAt, completedAt.
lastAttempt	The latest deploy attempt: id, phase (Running, Succeeded or Failed), startedAt, completedAt, summary and errors (each with recordedAt and message).

Example result

{
  "name": "web",
  "project": "proj-1a2b3c4d",
  "environment": "production",
  "state": "pending",
  "ready": false,
  "pendingChanges": 0,
  "replicas": { "desired": 2, "ready": 0, "available": 0 }
}

app_deployments

Lists a service’s deployment history, newest first. The ids are what rollback_app and deployment_logs take.

Argument	Type	Required	Meaning
app	string	yes	The service’s name.

• Needs: any token. Calls: GET /apps/{app}/deployments.
• Returns: {"items": [...]}. Each item has id, phase, image, commit, trigger, actor, changeTitle, createdAt, startedAt and completedAt. A rollback adds an entry of its own.

app_metrics

Shows CPU, memory, network and request use over time for a service, a database or a project.

Argument	Type	Required	Meaning
name	string	yes	The service, database or project to measure.
kind	string	no	apps (the default), databases or projects.
window	string	no	How far back: 15m, 1h, 6h or 24h. Default 1h: the tool sends 1h when you leave it out. (A raw call to the API with no window answers for 15m.)

• Needs: any token for apps (a project token for its own services). Any workspace token for databases and projects. Calls: GET /{kind}/{name}/metrics with window.
• Returns: window, step (seconds between points), available, an optional message, and series. Each series is a list of {"t": <unix seconds>, "v": <value>}. The series are cpu (cores), memory (bytes), network (bytes sent per second), networkIn (bytes received per second), requests (requests per second, services only) and storage (bytes, databases only). available is false, with a message, when Prometheus is not installed or not reachable (see Observability).

Example call

{ "name": "web", "window": "1h" }

Logs and variables

These three tools need a token that may read what a Read only token may not: Read only, with logs, Full access or a project token.

get_logs

Reads a service’s recent log lines. Use level or search to narrow them.

Argument	Type	Required	Meaning
app	string	yes	The service’s name.
tail	integer	no	How many of the newest lines to return in all. Default 100, at most 500. level and search look through the most recent lines only, so raise it to look further back.
since	string	no	Only lines newer than this long ago, such as 15m or 2h. Go duration syntax, up to 7 days.
level	string	no	Only this level: error, warn, info or debug.
search	string	no	Only lines containing this text, up to 200 characters.

• Needs: Read only, with logs, Full access, or a project token (for its own services). Calls: GET /apps/{app}/logs with tail, since, q and level.
• Returns: plain text, one line per log line, oldest first, <time in RFC 3339> <pod> <text>. When nothing matches: No log lines matched.
• Filters. The dashboard reads tail lines from the end of each copy’s log and applies level and search to those, so level error finds the errors among each copy’s newest lines. level warn also returns errors. A service with several copies returns the newest tail lines of all of them together.

Example result

2026-10-12T09:41:07Z web-6d9c7b8f5-x2k4p Error: connect ECONNREFUSED 203.0.113.10:5432

deployment_logs

Returns the build and deploy logs of one deployment. Read these when a deploy failed.

Argument	Type	Required	Meaning
app	string	yes	The service’s name.
deployment	string	yes	A deployment id from app_deployments.

• Needs: Read only, with logs, Full access, or a project token. Calls: GET /apps/{app}/deployments/{deployment}/logs.
• Returns: id and sections, a list of {"id", "title", "lines"} for the deploy attempt, the build (for a repository) and the rollout. Sections with nothing in them are left out. lines is the stage’s log as one block of text, a line to a line. A stage longer than 200 lines is cut to its last 200, which is where a failure says why, and starts with ... (N earlier lines left out). The whole log is at GET /apps/{app}/deployments/{deployment}/logs.

Example result

{
  "id": "web-7x2kq",
  "sections": [
    { "id": "provision", "title": "Provision", "lines": "Provisioning failed\n2026-10-12T09:41:07Z image pull failed" },
    { "id": "build", "title": "Build", "lines": "Kaniko completed without an immutable image digest" }
  ]
}

list_variables

Lists the names of a service’s variables and secrets.

Argument	Type	Required	Meaning
app	string	yes	The service’s name.

• Needs: Read only, with logs, Full access, or a project token. Calls: GET /apps/{app}/variables.
• Returns: secrets (a list of names), env (a list of {"name", "kind"}, where kind is literal or secret) and config (an object of plain configuration).

Secret values are never returned: secrets come back by name only. config is an object of plain configuration with its values, because plain configuration is not secret. The tool’s description says so. Store anything sensitive as a secret.

Example result

{
  "secrets": ["STRIPE_KEY"],
  "env": [],
  "config": { "LOG_LEVEL": "info" }
}

Databases, buckets and nodes

All of these are hidden from project tokens.

list_databases

Lists databases with their state.

Argument	Type	Required	Meaning
project	string	no	Only databases in this project.
environment	string	no	Only databases in this environment.

• Needs: any workspace token. Calls: GET /databases.
• Returns: {"items": [...]} of summaries like list_apps. details holds engine, highAvailability, and placement and version when set.

get_database

Shows one database’s settings and status.

Argument	Type	Required	Meaning
name	string	yes	The database’s name.

• Needs: any workspace token. Calls: GET /databases/{name}.
• Returns: {"resource": {...}, "spec": {...}, "status": {...}}. A token that cannot change things (Read only or Read only, with logs) gets the engine, host and connection details removed.

database_connection

Shows how to reach a database: host, port, database and user. The password and the connection string are never returned.

Argument	Type	Required	Meaning
name	string	yes	The database’s name.

• Needs: Full access. It is offered with --read-only too, since it only reads. Calls: GET /databases/{name}/connection, never with reveal.
• Returns: the connection facts that exist, among host, port, database and username.

Example result

{
  "database": "orders",
  "host": "orders-db.example.svc",
  "port": "5432",
  "username": "app"
}

list_buckets

Lists object storage buckets.

Argument	Type	Required	Meaning
project	string	no	Only buckets in this project.
environment	string	no	Only buckets in this environment.

• Needs: any workspace token. Calls: GET /object-stores.
• Returns: {"items": [...]} of summaries. A token that cannot change things does not get the bucket names or the endpoint.

list_nodes

Lists the cluster’s nodes with their region, zone and tags, including compliance tags.

• Arguments: none.
• Needs: any workspace token. Calls: GET /nodes.
• Returns: {"items": [...]}. Each node has name, ready, schedulable, controlPlane, region, zone, tags (each with name, and reserved and compliance when true), cpu and memory.

project_compliance

Shows the node tags a project requires, such as hipaa, and where it falls short of them.

Argument	Type	Required	Meaning
project	string	yes	The project’s ID.

• Needs: any workspace token. Calls: GET /projects/{project}/compliance.
• Returns: requiredTags, region, databases (each with name, compliant and missing) and openNodes, the nodes that carry a required tag without being reserved for it. See Compliance tags.

Deploying and changing services

These tools need Full access or a project token, and are absent with --read-only.

deploy_app

Deploys a service’s pending changes. With wait, waits until the service is ready or has failed and returns the outcome, including why a build failed.

Argument	Type	Required	Meaning
app	string	yes	The service to deploy.
wait	boolean	no	Wait until it is ready or has failed, and return the outcome (recommended).
timeout_seconds	integer	no	How long to wait. Default 300, at most 900.

• Calls: POST /apps/{app}/deploy, and with wait, GET /apps/{app}/status every 3 seconds.
• Returns without wait: {"app": "web", "deploying": true, "next": "call app_status to see how it is going"}.
• Returns with wait: the outcome.

Field	Meaning
app	The service.
status	The last status read, as returned by app_status.
ready	true only when the service is ready, has no pending changes and its latest build, if it has one, succeeded.
failed	true when the deploy failed. Only present then.
error	Why it failed (the deploy failed: <reason>), or why the wait ended (gave up waiting for web after 5m0s (it is pending): context deadline exceeded). Absent when it succeeded.

A failed deploy and a wait that ran out are results, not errors, so the agent reads the reason. If the agent’s client sends a progress token with the call, the server reports progress while it waits, as lines such as web: pending, build Running.

Example call

{ "app": "web", "wait": true, "timeout_seconds": 300 }

Example result

{
  "app": "web",
  "ready": true,
  "status": {
    "name": "web",
    "project": "proj-1a2b3c4d",
    "environment": "production",
    "state": "ready",
    "url": "https://web.example.com",
    "ready": true,
    "pendingChanges": 0,
    "replicas": { "desired": 2, "ready": 2, "available": 2 }
  }
}

Example result of a failed build

{
  "app": "web",
  "error": "the deploy failed: Kaniko completed without an immutable image digest",
  "failed": true,
  "ready": false,
  "status": {
    "name": "web",
    "project": "proj-1a2b3c4d",
    "environment": "production",
    "state": "pending",
    "ready": false,
    "pendingChanges": 0,
    "build": {
      "name": "web-build-x7k2p",
      "phase": "Failed",
      "failureReason": "Kaniko completed without an immutable image digest"
    }
  }
}

rollback_app

Returns a service to an earlier deployment.

Argument	Type	Required	Meaning
app	string	yes	The service to roll back.
deployment	string	no	The deployment to return to. The one before the newest in app_deployments by default.
wait	boolean	no	Wait until it is ready and return the outcome. It waits up to 300 seconds.

• Calls: GET /apps/{app}/deployments (without deployment), POST /apps/{app}/rollback, and with wait, status reads.
• Returns without wait: {"app": "web", "rollingBackTo": "web-abc12"}. With wait: the same outcome as deploy_app.
• It fails with <app> has no earlier deployment to go back to when there is only one deployment, and for a function (functions cannot be rolled back). A rollback records a deployment of its own, so a second call with no deployment goes to the entry before that one. Pass deployment to choose.

scale_app

Sets how many copies of a service run. It takes effect at once.

Argument	Type	Required	Meaning
app	string	yes	The service to scale.
replicas	integer	yes	How many copies should run, 0 or more. The API refuses more than 1000.

• Calls: POST /apps/{app}/scale.
• Returns: {"app": "web", "replicas": 4}.

set_sleep_when_idle

Makes a service stop when it has had no requests for a while and start again on the next request, or keeps it running. It needs the serverless add-on and a public host. The setting is saved at once, and the tool then waits up to 10 seconds for the cluster to say whether the service can sleep. See Scale to zero.

Argument	Type	Required	Meaning
app	string	yes	The service to change.
enabled	boolean	yes	true to stop it when idle and start it on the next request, false to keep it running.
idle_seconds	integer	no	How long without requests before it stops, 30 to 86400. The API’s default is 300. Anything else is refused with idle_seconds must be between 30 and 86400.

• Calls: PATCH /apps/{app} with {"scaleToZero": "on" | "off", "idleSeconds": ...}, then, when enabled is true, GET /apps/{app}/status until scaleToZero.active is true or the wait ends. A service that shows no request to sleep yet is waited for too, since the read can lag the change.
• Returns: {"app": "web", "sleepWhenIdle": true, "active": true} when the service sleeps when idle. When it cannot, because the add-on is missing, the service has no public host or CPU autoscaling is on, the setting stays saved, the service stays running, and the result says so: "active": false, a reason (the cluster’s message, or that it has not said yet) and a note. With enabled false: {"app": "web", "sleepWhenIdle": false}. When the setting was saved but reading the state back failed, the result is not an error: it has sleepWhenIdle and a note that says the check could not be made and why.

set_variable

Sets a variable or a secret on a service. It applies when the service is deployed.

Argument	Type	Required	Meaning
app	string	yes	The service.
key	string	yes	The variable’s name, such as LOG_LEVEL.
value	string	yes	Its value.
secret	boolean	no	Store it as a secret.
deploy	boolean	no	Apply it now by deploying, and wait up to 300 seconds for the outcome.

• Calls: POST /apps/{app}/config/set, or /secrets/set with secret; with deploy, POST /apps/{app}/deploy and status reads.
• Returns without deploy: {"app": "web", "change": "set", "key": "LOG_LEVEL", "next": "call deploy_app to apply it", "pending": true}. With deploy: app, change, key and deploy, which holds the outcome described under deploy_app.
• The result never repeats the value.

unset_variable

Removes a variable or a secret from a service. It applies when the service is deployed.

Argument	Type	Required	Meaning
app	string	yes	The service.
key	string	yes	The variable to remove.
secret	boolean	no	It is a secret.
deploy	boolean	no	Apply it now by deploying.

• Calls: POST /apps/{app}/config/delete, or /secrets/delete with secret; with deploy, a deploy and status reads.
• Returns: like set_variable, with "change": "removed". Removing a key that does not exist is not an error.

create_app

Creates a service from an image or a GitHub repository. It is saved as a draft unless deploy is set.

Argument	Type	Required	Meaning
project	string	yes	The project (its ID).
environment	string	yes	The environment, such as production.
name	string	yes	The service’s name.
image	string	no	A container image to run. Give this or repository.
repository	string	no	A GitHub repository to build, owner/name.
branch	string	no	The branch to build. Default main.
port	integer	no	The port it serves on.
deploy	boolean	no	Deploy it now instead of saving a draft.

• Needs: Full access. Hidden from project tokens. Calls: POST /apps with draft, or with deploy when the argument is set.
• Returns: {"created": "web", "draft": true}. With deploy: {"created": "web", "draft": false, "deploying": true, "next": "call app_status to see how it is going"}, and the service is deploying. The tool reports that only when the API says it deployed the service. A Geass older than the tool saves a draft instead, and the result then says "draft": true with a note that it did not deploy it: call deploy_app. With neither image nor repository the call fails with give an image or a repository to run. If you give both, image wins. A function cannot be created with this tool.
• With deploy, the deploy is checked before anything is saved: a cluster with no room for the service fails the call and leaves no draft behind. A repository also needs the project to be linked to GitHub first (see GitHub and builds).

Functions

These tools work on a service that runs a function (see Functions).

set_function_trigger

Calls a function on a schedule, adding the schedule or replacing the one with the same name. The schedule starts once the function is next deployed.

Argument	Type	Required	Meaning
app	string	yes	The function.
name	string	yes	The schedule’s name, such as nightly: lowercase letters, numbers and dashes, up to 32 characters.
schedule	string	to add	A cron expression in UTC with five fields, or @hourly, @daily, @weekly, @monthly, @yearly. The call fails with name and schedule are required without it.
path	string	no	The path to call. Must start with /. Default /.
payload	string	no	The JSON body to send. Must be valid JSON.

• Calls: GET /apps/{app}, then PATCH /apps/{app} with the full list of schedules. A function can have 8.
• Returns: {"app": "report", "next": "call deploy_app to start the schedule", "pending": true, "trigger": "nightly"}.

remove_function_trigger

Stops calling a function on a schedule. It stops once the function is next deployed.

Argument	Type	Required	Meaning
app	string	yes	The function.
name	string	yes	The schedule to remove.

• Calls: GET /apps/{app}, then PATCH /apps/{app}.
• Returns: {"app": "report", "pending": true, "removed": "nightly"}. A name that does not exist fails with <app> has no schedule named "<name>" and changes nothing.

invoke_function

Calls a function at its public address and returns its answer. This runs the function, which may change things, and wakes it if it is asleep.

Argument	Type	Required	Meaning
app	string	yes	The function to call.
path	string	no	The path to call, with a query if needed. Default /.
method	string	no	GET, or POST when there is a body (the default).
body	string	no	The JSON body to send.

• Calls: GET /apps/{app}, then an HTTP request to the function’s URL, without a Geass token. A call times out after 2 minutes, and at most 1 MiB of the answer is read.
• Returns: {"status": 200, "body": "..."}. A function that answers an error status is still a result.
• It fails with the function has no public address to call: give it a domain first for a function without a public address.

Deleting

delete_resource

Permanently deletes a service, database, bucket or logical database. This cannot be undone and deletes data.

Argument	Type	Required	Meaning
kind	string	yes	apps, databases, object-stores or logical-databases.
name	string	yes	The resource’s name.
confirm	string	yes	The resource’s name again, to confirm that it is to be deleted for good.

• Needs: Full access, and geass mcp --allow-delete. Without the flag the tool does not exist. Hidden from project tokens and from --read-only. Calls: DELETE /{kind}/{name}?confirmName={name}.
• Returns: {"deleted": "orders"}. It fails with not deleted: confirm must be exactly "orders" when confirm is not the name, and with kind must be apps, databases, object-stores or logical-databases for another kind.
• It is the only tool with destructiveHint: true, so a host that asks before destructive tools asks before this one.

Example call

{ "kind": "databases", "name": "orders", "confirm": "orders" }

Related

• MCP server for AI agents
• CLI commands
• Command line
• Users and API tokens
• REST API