5m-mcp

Everything agent_api exposes is convar-driven. All convars are read once at resource start; restart agent_api to pick up a change.

The sandbox-permission subset (readonly, write/control roots, rate limit, native blocklists, shell allowlist, plugin gates) can also be edited live from the web dashboard — no restart, persisted to dist/permissions.json.

Core

set agent_api_token                       ""          # blank = auto-generate
set agent_api_readonly                    false       # closes every mutating tool
set agent_api_root                        "resources/[agent]"
set agent_api_allow_write_paths           ""          # csv, extra write roots
set agent_api_allow_control_paths         ""          # csv, extra lifecycle roots

convar	default	meaning
`agent_api_token`	(blank)	bearer token in `x-agent-token` header. Blank → generate + persist to `dist/.agent_token`
`agent_api_readonly`	`false`	when `true`, every mutating tool (write, ensure/start/..., shell, ESX add_money, native verbs not starting with Get/Has/Is/...) refuses
`agent_api_root`	`resources/[agent]`	sandbox root for writes + lifecycle
`agent_api_allow_write_paths`	(none)	additional write roots (csv)
`agent_api_allow_control_paths`	(none)	additional lifecycle roots (csv)

Buffers and limits

set agent_api_console_buffer_lines        2000
set agent_api_max_file_bytes              2097152     # 2 MB
set agent_api_extra_write_extensions      ""          # csv, e.g. ".webp,.lock"
set agent_api_rate_per_minute             120

convar	default	meaning
`agent_api_console_buffer_lines`	`2000`	size of the in-memory ring buffer used by `tail_console`
`agent_api_max_file_bytes`	`2_097_152`	hard cap on `read_file` / `write_file`
`agent_api_extra_write_extensions`	(none)	append extensions to the read + write allowlist (csv, with or without the leading `.`)
`agent_api_rate_per_minute`	`120`	token-bucket cap per `x-agent-token` hash; over → `429 RATE_LIMITED`

Live client testing

set agent_api_test_session_ttl_seconds    1800
set agent_api_test_max_subjects           4
set agent_api_client_blocked_natives      ""          # csv

convar	default	meaning
`agent_api_test_session_ttl_seconds`	`1800`	how long `/agent_test_optin` stays valid
`agent_api_test_max_subjects`	`4`	size of the active subject pool
`agent_api_client_blocked_natives`	(none)	per-name blocklist enforced before `client_call_native`

Server-side natives

set agent_api_server_blocked_natives      ""          # csv

In addition to the user blocklist, server_call_native carries a built-in danger list that always refuses: DropPlayer, ExecuteCommand, StopResource, StartResource, ScheduleResourceTick, PrintStructuredTrace, CancelEvent, TempBanPlayer, BanPlayer. Use the dedicated tools (run_command, ensure_resource, etc.) for those instead.

Shell

set agent_api_shell_allowed_commands      ""          # csv; default = npm,npx,pnpm,yarn,bun,vite,git,node
set agent_api_node_binary                 "node"      # used by screenshot_nui spawn

If agent_api_shell_allowed_commands is non-empty it replaces the default — set it to your full allowlist, not just additions.

Plugins

set agent_api_plugin_esx_enabled          auto        # auto | true | false
set agent_api_plugin_oxlib_enabled        auto
set agent_api_plugin_oxmysql_enabled      auto

auto — load only if the target resource is started. Default.
true / force — load even if detection fails.
false — never load, even if detected.

Plugin-specific

set agent_api_plugin_esx_blocked_methods           ""          # csv of xPlayer/ESX method names to refuse
set agent_api_plugin_oxlib_blocked_methods         ""          # csv of ox_lib server export names to refuse
set agent_api_plugin_oxmysql_readonly              true        # SELECT-only
set agent_api_plugin_oxmysql_allow_statements      "SELECT"    # csv, uppercase, e.g. "SELECT,INSERT,UPDATE"

Full example `server.cfg` block

# agent_api
set agent_api_token                       ""
set agent_api_readonly                    false
set agent_api_root                        "resources/[agent]"
set agent_api_rate_per_minute             120
set agent_api_test_session_ttl_seconds    1800

# Open oxmysql for full CRUD on app-owned tables only
set agent_api_plugin_oxmysql_readonly         false
set agent_api_plugin_oxmysql_allow_statements "SELECT,INSERT,UPDATE,DELETE"

# ACE
add_ace resource.agent_api command.ensure  allow
add_ace resource.agent_api command.start   allow
add_ace resource.agent_api command.stop    allow
add_ace resource.agent_api command.restart allow
add_ace resource.agent_api command.refresh allow
add_ace resource.agent_api command.say     allow

ensure agent_api

Configuration