Configuration Schema

JSON Schema for the Operator configuration file (config.toml).

Schema Information

$schema: https://json-schema.org/draft/2020-12/schema
title: Config

Required Fields

agents
notifications
queue
paths
ui
launch
templates

Properties

Property	Type	Required	Description
`projects`	`array`	No	List of projects operator can assign work to
`agents`	→ `AgentsConfig`	Yes
`notifications`	→ `NotificationsConfig`	Yes
`queue`	→ `QueueConfig`	Yes
`paths`	→ `PathsConfig`	Yes
`ui`	→ `UiConfig`	Yes
`launch`	→ `LaunchConfig`	Yes
`templates`	→ `TemplatesConfig`	Yes
`api`	→ `ApiConfig`	No
`logging`	→ `LoggingConfig`	No
`tmux`	→ `TmuxConfig`	No
`sessions`	→ `SessionsConfig`	No	Session wrapper configuration (tmux, vscode, or cmux)
`llm_tools`	→ `LlmToolsConfig`	No
`rest_api`	→ `RestApiConfig`	No
`git`	→ `GitConfig`	No
`kanban`	→ `KanbanConfig`	No	Kanban provider configuration for syncing issues from Jira, Linear, etc.
`version_check`	→ `VersionCheckConfig`	No	Version check configuration for automatic update notifications
`delegators`	`array`	No	Agent delegator configurations for autonomous ticket launching
`model_servers`	`array`	No	User-declared model servers (ollama, lmstudio, any OpenAI-compat host). Implicit builtin servers exist for each `llm_tool`’s vendor API and do not need declaration.
`relay`	→ `RelayConfig`	No	Relay MCP injection configuration
`mcp`	→ `McpConfig`	No	Model Context Protocol (MCP) server configuration
`acp`	→ `AcpConfig`	No	Agent Client Protocol (ACP) agent configuration

Type Definitions

AgentsConfig

Property	Type	Required	Description
`max_parallel`	`integer`	Yes
`cores_reserved`	`integer`	Yes
`max_agents_per_repo`	`integer`	No	Maximum concurrent agents per project/repo (default: 1). Requires `git.use_worktrees` = true when > 1 to avoid conflicts.
`health_check_interval`	`integer`	Yes
`generation_timeout_secs`	`integer`	No	Timeout in seconds for each agent generation (default: 300 = 5 min)
`sync_interval`	`integer`	No	Interval in seconds between ticket-session syncs (default: 60)
`step_timeout`	`integer`	No	Maximum seconds a step can run before timing out (default: 1800 = 30 min)
`silence_threshold`	`integer`	No	Seconds of tmux silence before considering agent awaiting input (default: 30)

NotificationsConfig

Notifications configuration with support for multiple integrations.

Property	Type	Required	Description
`enabled`	`boolean`	Yes	Global enabled flag for all notifications
`os`	→ `OsNotificationConfig`	No	OS notification configuration
`webhook`	object	No	Single webhook configuration (for simple setups)
`webhooks`	`array`	No	Multiple webhook configurations

OsNotificationConfig

OS notification configuration.

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether OS notifications are enabled
`sound`	`boolean`	No	Play sound with notifications
`events`	`array`	No	Events to send (empty = all events) Possible values: agent.started, agent.completed, agent.failed, `agent.awaiting_input`, `agent.session_lost`, pr.created, pr.merged, pr.closed, `pr.ready_to_merge`, `pr.changes_requested`, ticket.returned, investigation.created

WebhookConfig

Webhook notification configuration.

Property	Type	Required	Description
`name`	`string` \| `null`	No	Optional name for this webhook (for logging)
`enabled`	`boolean`	No	Whether this webhook is enabled
`url`	`string`	No	Webhook URL
`auth_type`	`string` \| `null`	No	Authentication type: “bearer” or “basic”
`token_env`	`string` \| `null`	No	Environment variable containing the bearer token
`username`	`string` \| `null`	No	Username for basic auth
`password_env`	`string` \| `null`	No	Environment variable containing the password for basic auth
`events`	`array` \| `null`	No	Events to send (empty = all events)

QueueConfig

Property	Type	Required
`auto_assign`	`boolean`	Yes
`priority_order`	`array`	Yes
`poll_interval_ms`	`integer`	Yes

PathsConfig

Property	Type	Required	Description
`tickets`	`string`	Yes
`projects`	`string`	Yes
`state`	`string`	Yes
`worktrees`	`string`	No	Base directory for per-ticket worktrees (default: ~/.operator/worktrees)

UiConfig

Property	Type	Required
`refresh_rate_ms`	`integer`	Yes
`completed_history_hours`	`integer`	Yes
`summary_max_length`	`integer`	Yes
`panel_names`	→ `PanelNamesConfig`	No

PanelNamesConfig

Property	Type	Required
`status`	`string`	No
`queue`	`string`	No
`in_progress`	`string`	No
`completed`	`string`	No

LaunchConfig

Property	Type	Required	Description
`confirm_autonomous`	`boolean`	Yes
`confirm_paired`	`boolean`	Yes
`launch_delay_ms`	`integer`	Yes
`docker`	→ `DockerConfig`	No	Docker execution configuration
`yolo`	→ `YoloConfig`	No	YOLO (auto-accept) mode configuration

DockerConfig

Docker execution configuration for running agents in containers

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether docker mode option is available in launch dialog
`image`	`string`	No	Docker image to use (required if enabled)
`extra_args`	`array`	No	Additional docker run arguments
`mount_path`	`string`	No	Container mount path for the project (default: /workspace)
`env_vars`	`array`	No	Environment variables to pass through to the container

YoloConfig

YOLO (auto-accept) mode configuration for fully autonomous execution

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether YOLO mode option is available in launch dialog

TemplatesConfig

Property	Type	Required	Description
`preset`	→ `CollectionPreset`	No	Named preset for issue type collection Options: simple, `dev_kanban`, `devops_kanban`, custom
`collection`	`array`	No	Custom issuetype collection (only used when preset = custom) List of issue type keys: TASK, FEAT, FIX, SPIKE, INV
`active_collection`	`string` \| `null`	No	Active collection name (overrides preset if set) Can be a builtin preset name or a user-defined collection

CollectionPreset

Predefined issue type collections

Allowed Values:

simple - Simple tasks only: TASK
dev_kanban - Developer kanban: TASK, FEAT, FIX
devops_kanban - DevOps kanban: TASK, SPIKE, INV, FEAT, FIX
custom - Custom collection (use the collection field)

ApiConfig

API integrations configuration

Property	Type	Required	Description
`pr_check_interval_secs`	`integer`	No	Interval in seconds between PR status checks (default: 60)
`rate_limit_check_interval_secs`	`integer`	No	Interval in seconds between rate limit checks (default: 300)
`rate_limit_warning_threshold`	`number`	No	Show warning when rate limit remaining is below this percentage (default: 0.2)

LoggingConfig

Logging configuration

Property	Type	Required	Description
`level`	`string`	No	Log level filter (trace, debug, info, warn, error)
`to_file`	`boolean`	No	Whether to log to file in TUI mode (false = stderr for debugging)

TmuxConfig

Property	Type	Required	Description
`config_generated`	`boolean`	No	Whether custom tmux config has been generated

SessionsConfig

Session wrapper configuration

Controls how operator creates and manages terminal sessions for agents. Four modes are supported:

tmux: Standalone tmux sessions (default)
vscode: VS Code integrated terminal (requires extension)
cmux: macOS terminal multiplexer (requires running inside cmux)
zellij: Zellij terminal workspace manager

Property	Type	Required	Description
`wrapper`	→ `SessionWrapperType`	No	Which session wrapper to use
`tmux`	→ `SessionsTmuxConfig`	No	Tmux-specific configuration
`vscode`	→ `SessionsVSCodeConfig`	No	VS Code-specific configuration
`cmux`	→ `SessionsCmuxConfig`	No	cmux-specific configuration
`zellij`	→ `SessionsZellijConfig`	No	Zellij-specific configuration

SessionWrapperType

Session wrapper type for terminal session management

Allowed Values:

tmux - Standalone tmux sessions (default)
vscode - VS Code integrated terminal (via extension webhook)
cmux - cmux macOS terminal multiplexer
zellij - Zellij terminal workspace manager

SessionsTmuxConfig

Tmux-specific session configuration

Property	Type	Required	Description
`config_generated`	`boolean`	No	Whether custom tmux config has been generated
`socket_name`	`string`	No	Socket name for session isolation

SessionsVSCodeConfig

VS Code extension session configuration

Property	Type	Required	Description
`webhook_port`	`integer`	No	Port for extension webhook server
`connect_timeout_ms`	`integer`	No	Connection timeout in milliseconds

SessionsCmuxConfig

cmux macOS terminal multiplexer session configuration

Property	Type	Required	Description
`binary_path`	`string`	No	Path to the cmux binary
`require_in_cmux`	`boolean`	No	Require running inside cmux (`CMUX_WORKSPACE_ID` env var present)
`placement`	→ `CmuxPlacementPolicy`	No	Where to place new agent sessions: “auto”, “workspace”, or “window”

CmuxPlacementPolicy

Placement policy for cmux sessions: where to create new agent terminals

Allowed Values:

auto - Automatically choose: 0-1 windows → new workspace, >1 windows → new window
workspace - Always create a new workspace in the active window
window - Always create a new window for each ticket

SessionsZellijConfig

Zellij terminal workspace manager session configuration

Property	Type	Required	Description
`require_in_zellij`	`boolean`	No	Require running inside Zellij (ZELLIJ env var present)

LlmToolsConfig

LLM CLI tools configuration

Property	Type	Required	Description
`detected`	`array`	No	Detected CLI tools (populated on first startup)
`providers`	`array`	No	Available {tool, model} pairs for launching tickets Built from detected tools + their model aliases
`detection_complete`	`boolean`	No	Whether detection has been completed
`default_tool`	`string` \| `null`	No	User’s preferred default LLM tool (e.g., “claude”)
`default_model`	`string` \| `null`	No	User’s preferred default model alias (e.g., “opus”)
`skill_directory_overrides`	`object`	No	Per-tool overrides for skill directories (keyed by `tool_name`)

DetectedTool

A detected CLI tool (e.g., claude binary)

Property	Type	Required	Description
`name`	`string`	Yes	Tool name (e.g., “claude”)
`path`	`string`	Yes	Path to the binary
`version`	`string`	Yes	Version string
`min_version`	`string` \| `null`	No	Minimum required version for Operator compatibility
`version_ok`	`boolean`	No	Whether the installed version meets the minimum requirement
`model_aliases`	`array`	No	Available model aliases (e.g., [“opus”, “sonnet”, “haiku”])
`command_template`	`string`	No	Command template with , , placeholders
`capabilities`	→ `ToolCapabilities`	No	Tool capabilities
`yolo_flags`	`array`	No	CLI flags for YOLO (auto-accept) mode

ToolCapabilities

Tool capabilities

Property	Type	Required	Description
`supports_sessions`	`boolean`	No	Whether the tool supports session continuity via UUID
`supports_headless`	`boolean`	No	Whether the tool can run in headless/non-interactive mode

LlmProvider

A {tool, model} pair that can be selected when launching tickets. Includes optional variant fields adopted from vibe-kanban’s profile system.

Property	Type	Required	Description
`tool`	`string`	Yes	CLI tool name (e.g., “claude”, “codex”, “gemini”)
`model`	`string`	Yes	Model alias or name (e.g., “opus”, “sonnet”, “gpt-4.1”)
`display_name`	`string` \| `null`	No	Optional display name for UI (e.g., “Claude Opus”, “Codex High”)
`flags`	`array`	No	Additional CLI flags for this provider (e.g., [”–dangerously-skip-permissions”])
`env`	`object`	No	Environment variables to set when launching
`approvals`	`boolean`	No	Whether this provider requires approval gates
`plan_only`	`boolean`	No	Whether to run in plan-only mode
`reasoning_effort`	`string` \| `null`	No	Reasoning effort level (Codex: “low”, “medium”, “high”)
`sandbox`	`string` \| `null`	No	Sandbox mode (Codex: “danger-full-access”, “workspace-write”)

SkillDirectoriesOverride

Per-tool skill directory overrides

Property	Type	Required	Description
`global`	`array`	No	Additional global skill directories
`project`	`array`	No	Additional project-relative skill directories

RestApiConfig

REST API server configuration

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether the REST API is enabled
`host`	`string`	No	Address the REST API binds to. Defaults to `127.0.0.1` (local only) so the server — which reports the project directory name — is not reachable from other hosts. Set to `0.0.0.0` to expose it on all interfaces.
`port`	`integer`	No	Port for the REST API server
`cors_origins`	`array`	No	CORS allowed origins (empty = allow all)

GitConfig

Git provider configuration for PR/MR operations

Property	Type	Required	Description
`provider`	object	No	Active provider (auto-detected from remote URL if not specified)
`github`	→ `GitHubConfig`	No	GitHub-specific configuration
`gitlab`	→ `GitLabConfig`	No	GitLab-specific configuration (planned)
`branch_format`	`string`	No	Branch naming format (e.g., “{type}/{ticket_id}-{slug}”)
`use_worktrees`	`boolean`	No	Whether to use git worktrees for per-ticket isolation (default: false) When false, tickets work directly in the project directory with branches

GitProviderConfig

Git provider selection

Allowed Values:

github - GitHub (github.com)
gitlab - GitLab (gitlab.com or self-hosted)
bitbucket - Bitbucket (bitbucket.org)
azuredevops - Azure DevOps (dev.azure.com)

GitHubConfig

GitHub-specific configuration

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether GitHub integration is enabled
`token_env`	`string`	No	Environment variable containing the GitHub token (default: `GITHUB_TOKEN`)

GitLabConfig

GitLab-specific configuration (planned)

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether GitLab integration is enabled
`token_env`	`string`	No	Environment variable containing the GitLab token (default: `GITLAB_TOKEN`)
`host`	`string` \| `null`	No	GitLab host (default: gitlab.com, can be self-hosted)

KanbanConfig

Kanban provider configuration for syncing issues from external systems

Providers are keyed by domain/workspace:

Jira: keyed by domain (e.g., “foobar.atlassian.net”)
Linear: keyed by workspace slug (e.g., “myworkspace”)
GitHub Projects: keyed by owner login (e.g., “my-org”)

Property	Type	Required	Description
`jira`	`object`	No	Jira Cloud instances keyed by domain (e.g., “foobar.atlassian.net”)
`linear`	`object`	No	Linear instances keyed by workspace slug
`github`	`object`	No	GitHub Projects v2 instances keyed by owner login (user or org) NOTE: This is the kanban GitHub integration (Projects v2), distinct from `GitHubConfig` which is the git provider used for PRs and branches. The two use different env vars and different scopes — see `docs/getting-started/kanban/github.md` for the full disambiguation.

JiraConfig

Jira Cloud provider configuration

The domain is specified as the HashMap key in KanbanConfig.jira

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether this provider is enabled
`api_key_env`	`string`	No	Environment variable name containing the API key (default: `OPERATOR_JIRA_API_KEY`)
`email`	`string`	No	Atlassian account email for authentication
`projects`	`object`	No	Per-project sync configuration

ProjectSyncConfig

Per-project/team sync configuration for a kanban provider

Property	Type	Required	Description
`sync_user_id`	`string`	No	User ID to sync issues for (provider-specific format) - Jira: accountId (e.g., “5e3f7acd9876543210abcdef”) - Linear: user ID (e.g., “abc12345-6789-0abc-def0-123456789abc”) - GitHub Projects: numeric GitHub `databaseId` (e.g., “12345678”)
`sync_statuses`	`array`	No	Workflow statuses to sync (empty = default/first status only)
`collection_name`	`string` \| `null`	No	Optional `IssueTypeCollection` name this project maps to. Not required for kanban onboarding or sync.
`type_mappings`	`object`	No	Explicit mapping: kanban issue type ID → operator issue type key (e.g., TASK, FEAT, FIX). Multiple kanban types can map to the same operator template.
`bidirectional`	`boolean`	No	When true, operator pushes status changes and activity logs back to this kanban project. Ticket state changes (todo→doing, doing→done) and step completions with delegator info are reflected upstream. Default: false.

LinearConfig

Linear provider configuration

The workspace slug is specified as the HashMap key in KanbanConfig.linear

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether this provider is enabled
`api_key_env`	`string`	No	Environment variable name containing the API key (default: `OPERATOR_LINEAR_API_KEY`)
`projects`	`object`	No	Per-team sync configuration

GithubProjectsConfig

GitHub Projects v2 (kanban) provider configuration

The owner login (user or org) is specified as the HashMap key in KanbanConfig.github. Project keys inside projects are GraphQL node IDs (e.g., PVT_kwDOABcdefg) — opaque, stable identifiers used directly by every GitHub Projects v2 mutation without needing a lookup.

Distinct from GitHubConfig (the git provider used for PR/branch operations). They live in different parts of the config tree, use different env vars (OPERATOR_GITHUB_TOKEN vs GITHUB_TOKEN), and require different OAuth scopes (project vs repo). See docs/getting-started/kanban/github.md for the full rationale.

Property	Type	Required	Description
`enabled`	`boolean`	No	Whether this provider is enabled
`api_key_env`	`string`	No	Environment variable name containing the GitHub token (default: `OPERATOR_GITHUB_TOKEN`). The token must have `project` (or `read:project`) scope, NOT just `repo` — see the disambiguation guide in the kanban github docs.
`projects`	`object`	No	Per-project sync configuration. Keys are `GraphQL` project node IDs.

VersionCheckConfig

Version check configuration for automatic update notifications

Property	Type	Required	Description
`enabled`	`boolean`	No	Enable automatic version checking on startup
`url`	`string` \| `null`	No	URL to fetch latest version from (optional, can be removed)
`timeout_secs`	`integer`	No	Timeout in seconds for version check HTTP request

Delegator

Agent delegator configuration for autonomous ticket launching

A delegator is a named {tool, model} pairing with optional launch configuration that can be used to launch agents for tickets.

Property	Type	Required	Description
`name`	`string`	Yes	Unique name for this delegator (e.g., “claude-opus-auto”)
`llm_tool`	`string`	Yes	LLM tool name (must match a detected tool, e.g., “claude”, “codex”)
`model`	`string`	Yes	Model alias (e.g., “opus”, “sonnet”, “gpt-4o”)
`display_name`	`string` \| `null`	No	Optional display name for UI
`model_properties`	`object`	No	Arbitrary model properties (e.g., `reasoning_effort`, sandbox)
`launch_config`	object	No	Optional launch configuration
`model_server`	`string` \| `null`	No	Name of a declared `ModelServer` (from `Config.model_servers`). `None` means use the `llm_tool`’s implicit vendor default (claude → anthropic-api, codex → openai-api, gemini → google-api).
`remote_agent`	object	No	Declarative reference to a remote, named agent on another platform (e.g. an AGNT agent or an `OpenAI` Assistant; see [`crate::config::AgentProfile`]). Export-only: Operator has no runtime client for those platforms, so a delegator carrying this CANNOT be launched locally — resolution errors out (see `delegator_resolution`). It is stored, listed, serialized into an `AgentProfile`, and — for `platform == "agnt"` — surfaced in the `--format agnt` workflow export as an `agnt-agent` node. `None` = ordinary, locally launchable delegator.
`x_agnt`	object	No	Opaque AGNT-namespaced extension fields, preserved verbatim across an `AgentProfile` round-trip so re-export is lossless (e.g. `memory`, `assignedWorkflows`, `creditLimit`). Operator never interprets this.
`x_openai`	object	No	Opaque OpenAI-namespaced extension fields, preserved verbatim across an `AgentProfile` round-trip (e.g. `instructions`, `tools`, `tool_resources`, `metadata`, thread refs). Mirror of [`Self::x_agnt`]; never interpreted.
`unmapped_core`	object	No	Opaque carry for `AgentProfile` shared-core fields Operator cannot model first-class (`system_prompt` / `skills` / `mcp_servers` / `tools`) so an import→export round-trip is lossless. Distinct from `x_agnt`: these are shared-core fields, not AGNT-specific, so folding them into `x_agnt` would corrupt that namespace. Operator never interprets this.

DelegatorLaunchConfig

Launch configuration for a delegator

Controls how the delegator launches agents. Optional fields use tri-state semantics: None = inherit from global config, Some(true/false) = override.

Property	Type	Required	Description
`yolo`	`boolean`	No	Run in YOLO (auto-accept) mode
`permission_mode`	`string` \| `null`	No	Permission mode override
`flags`	`array`	No	Additional CLI flags
`use_worktrees`	`boolean` \| `null`	No	Override global `git.use_worktrees` per-delegator (None = use global setting)
`create_branch`	`boolean` \| `null`	No	Whether to create a git branch for the ticket (None = default behavior)
`docker`	`boolean` \| `null`	No	Run in docker container (None = use global `launch.docker.enabled`)
`prompt_prefix`	`string` \| `null`	No	Prompt text to prepend before the generated step prompt
`prompt_suffix`	`string` \| `null`	No	Prompt text to append after the generated step prompt
`operator_relay`	`boolean` \| `null`	No	Override global relay auto-inject MCP setting per-delegator (None = use global setting)

RemoteAgentRef

A declarative reference to a remote, named agent hosted by another platform.

platform is the hosting service ("agnt", "openai") — deliberately distinct from the core provider/llm_tool (the model or coding CLI). These agents are API/memory-native and live on the remote side; Operator has no runtime client for them, so a delegator carrying one is export-only and cannot be launched locally (see the guard in delegator_resolution).

Property	Type	Required	Description
`platform`	`string`	Yes	Hosting platform (e.g. `"agnt"`, `"openai"`).
`id`	`string`	Yes	Platform-native agent identifier (e.g. an AGNT agent name, an `OpenAI` `asst_…` id).

ModelServer

A named host that serves models via an inference API.

Model servers are orthogonal to llm_tools: a delegator pairs an agentic CLI (llm_tool, e.g. claude/codex/gemini) with a model-serving endpoint (model_server, e.g. ollama-local, openai-api, a custom vllm host).

Implicit builtin servers (anthropic-api, openai-api, google-api) are returned by [implicit_model_server_for_tool] and do not need to be declared in config.

Property	Type	Required	Description
`name`	`string`	Yes	Unique name (e.g., “ollama-local”, “vllm-gpu1”)
`kind`	`string`	Yes	Kind: “ollama”, “openrouter”, “openai-compat”, “anthropic-api”, “openai-api”, “google-api”, “lmstudio”
`base_url`	`string` \| `null`	No	Base URL of the inference endpoint (e.g., `http://localhost:11434`). `None` for implicit vendor servers means use the SDK default.
`api_key_env`	`string` \| `null`	No	Name of an env var providing the API key (e.g., `OLLAMA_API_KEY`)
`extra_env`	`object`	No	Additional environment variables set when spawning agents that use this server
`display_name`	`string` \| `null`	No	Optional display name for UI

RelayConfig

Relay MCP injection configuration

Property	Type	Required	Description
`auto_inject_mcp`	`boolean`	No	When true, automatically inject the relay MCP server for all delegators. When false (default), relay injection is opt-in per delegator.

McpConfig

Model Context Protocol (MCP) server configuration

Property	Type	Required	Description
`http_enabled`	`boolean`	No	Whether to mount MCP HTTP/SSE endpoints on the REST API server. Toggling requires an API restart (no hot-swap of the axum router).
`stdio_advertised`	`boolean`	No	Whether the descriptor endpoint advertises the `operator mcp` stdio command. Set to false on multi-tenant/remote deployments where clients shouldn’t spawn local subprocesses.
`expose_ticket_write_tools`	`boolean`	No	Whether to expose ticket-mutating tools (claim, complete, return-to-queue, create) over MCP. Defaults to `false` because any MCP client can call them.
`external_servers`	`array`	No	External MCP servers to inject into spawned agent sessions. Each entry produces a separate `--mcp-config` file alongside the relay config when launching Claude Code agents.

ExternalMcpServer

An external MCP server to inject into spawned agent sessions.

Values in command, args, and env support ${VAR} interpolation, expanded at spawn time from the operator process environment.

When discover_from is set, operator reads an MCP server spec from that JSON sidecar file at spawn time. The sidecar must contain a top-level mcpServer object with command, args, and env fields. If the file is absent and command is empty, the server is silently skipped.

Property	Type	Required	Description
`name`	`string`	Yes	Server name used as the key in the `mcpServers` JSON object (e.g., “kanbots”). Must be unique across all external servers.
`command`	`string`	No	Command to execute. Supports `${VAR}` interpolation.
`args`	`array`	No	Command arguments. Each element supports `${VAR}` interpolation.
`env`	`object`	No	Environment variables passed to the MCP server process. Values support `${VAR}` interpolation.
`enabled`	`boolean`	No	Whether this server is enabled. Allows disabling without removing config.
`discover_from`	`string` \| `null`	No	Path to a JSON sidecar discovery file. Relative paths resolve from the project directory. The sidecar must contain `{ "mcpServer": { ... } }`. When the file exists, its `mcpServer` spec is used verbatim (overriding `command`/`args`/`env`). When absent and `command` is empty, the server is silently skipped.

AcpConfig

Agent Client Protocol (ACP) agent configuration.

Operator runs as an ACP agent over stdio when editors (Zed, JetBrains, Emacs agent-shell, Kiro, etc.) spawn operator acp. Each ACP session maps to an in-progress ACP ticket and a delegator subprocess.

Property	Type	Required	Description
`stdio_advertised`	`boolean`	No	Whether the dashboard advertises the `operator acp` stdio entrypoint (and editor-config snippet actions). Set to false on machines that shouldn’t be used as ACP agents.
`default_delegator`	`string` \| `null`	No	Name of the delegator (from `[[delegators]]`) to use for ACP prompts. If unset or not found, falls back to the operator’s default delegator resolution.
`max_concurrent_sessions`	`integer`	No	Maximum number of concurrent ACP sessions. New `session/new` requests beyond this limit are rejected with a JSON-RPC error.