MCP server - Tinycloud

pnpm mcp

pnpm mcp starts the MCP stdio server. It uses TINY_API_URL / TINY_TOKEN or the current CLI profile. The server uses the official MCP SDK for protocol negotiation, request validation, cancellation, and lifecycle handling, and accepts both current and legacy MCP clients.

Connect a client

Point your MCP host at the repository’s mcp script. Replace the path below with your checkout path:

{
  "mcpServers": {
    "tinycloud": {
      "command": "pnpm",
      "args": ["--dir", "/path/to/tinycloud", "mcp"],
      "env": {
        "TINY_API_URL": "https://tinycloud.example.com",
        "TINY_TOKEN": "tiny_..."
      }
    }
  }
}

Omit env to reuse the profile created by tiny login. Keep stdout reserved for MCP messages; operational diagnostics are written to stderr.

The MCP server is a remote client of the same REST API the CLI uses. It never starts a local runtime, and it cannot do anything a human with the same token could not.

Read-only tools

Tool	Purpose
`tiny_get_platform_capabilities`	what this target actually supports — call this first
`tiny_list_capability_operations`	the broker operation catalog
`tiny_validate_manifest`	parse, normalize, and validate
`tiny_plan_deployment`	non-mutating plan with diffs and required approvals
`tiny_get_deployment`	state and failure detail
`tiny_get_logs`	bounded runtime logs
`tiny_list_apps`	apps in the organization
`tiny_get_app`	app detail
`tiny_list_deployments`	recent deployments for an app
`tiny_list_environments`	production, named, and preview environments
`tiny_list_access`	access bindings
`tiny_get_lifecycle_recommendations`	idle / archive / delete recommendations
`tiny_get_audit_events`	audit query
`tiny_get_usage`	usage samples
`tiny_list_webhooks`	webhook endpoints, without secrets

Mutating tools

Tool	Purpose
`tiny_deploy_app`	enqueue a deployment
`tiny_approve_deployment`	resume one paused in `awaiting_approval`
`tiny_rollback_deployment`	repoint traffic
`tiny_grant_access` / `tiny_revoke_access`	bindings
`tiny_request_capability`	request a grant
`tiny_approve_capability_grant`	approve one, as an org admin
`tiny_create_preview` / `tiny_delete_preview`	preview environments
`tiny_archive_app` / `tiny_restore_app`	lifecycle
`tiny_delete_app`	schedule deletion
`tiny_create_webhook` / `tiny_delete_webhook`	webhook endpoints

Organization tools

Tool	Purpose
`tiny_create_organization`	bootstrap
`tiny_add_member`	add a member with a role and groups
`tiny_create_connection`	store a vendor credential
`tiny_set_secret`	create a secret version
`tiny_list_secrets`	names and creation times only

Errors are results, not transport failures

Tool errors are returned as structured tool results with stable codes, remediation, retryability, and next actions — rather than as transport failures an assistant has to interpret from prose.

{
  "status": "approval_required",
  "reasons": [
    { "code": "NEW_PRODUCTION_CAPABILITY", "detail": "stripe.payments.read" }
  ],
  "next_actions": ["tiny_approve_deployment"]
}

Recommended sequence

tiny_get_platform_capabilities

Generate a manifest the target can run, rather than discovering it mid-deploy.

tiny_validate_manifest

tiny_plan_deployment

tiny_deploy_app, then poll tiny_get_deployment

Deploy is asynchronous.

tiny_get_logs on failure

The workflow guide is Deploying from an AI assistant.

​Connect a client

​Read-only tools

​Mutating tools

​Organization tools

​Errors are results, not transport failures

​Recommended sequence

Connect a client

Read-only tools

Mutating tools

Organization tools

Errors are results, not transport failures

Recommended sequence