Skip to content

MCP tools

These are the tools the agentry MCP server exposes to your harness. The agent calls them — you don't type them. But knowing what they do helps you prompt well and understand what the agent is doing on your behalf.

Your harness's tools/list is always the authoritative, up-to-date inventory; this page is the friendly tour.

Sandbox lifecycle

ToolWhat it does
sandbox_createSpin up a fresh, isolated sandbox. Returns the sandbox ID and the services already bound to it.
sandbox_listList sandboxes, or show details of one (status only — not for choosing where to work).
sandbox_deleteTear a sandbox down.

When sandbox_create returns, it tells the agent which services are already wired in (e.g. a database, an AI key) so the agent uses the real thing instead of stubbing a fake.

Running commands

ToolWhat it does
command_runRun a command and wait for it to finish (installs, tests, git, curl).
command_startStart a command in the background; returns an ID.
command_logsRead new output from a background command.
command_interruptStop a background command.

Long-running projects

For dev servers, watchers, schedulers — anything meant to keep running.

ToolWhat it does
project_createScaffold a managed project (Next.js, static HTML, Streamlit, FastAPI, a Python script, or custom).
project_startStart (or restart) the project.
project_stopStop it.
project_listList projects with status, ports, uptime, and last error.
project_logsRead recent logs, with an optional text filter.

Files

ToolWhat it does
file_readRead a file (optionally a line range).
file_writeCreate or overwrite a file (makes parent folders automatically).
file_listList files under a folder, optionally recursive and filtered.
file_grepSearch across many files with a regex.
file_replaceReplace an exact string in one file.
file_multi_editApply several edits to one file in a single step.

Ports

ToolWhat it does
port_waitWait until a port is accepting connections — used after starting a dev server.

Services & secrets

ToolWhat it does
service_listList services that can be bound (Postgres, Redis, MySQL, MongoDB, S3, SMTP, Stripe, OpenAI, Anthropic, and more).
service_bindWire a service into the sandbox; returns the env var names it set.
secret_setSet a non-secret env var the app needs (rejects secret-shaped values — use agentry env set for real secrets).
secret_listList env var names in the sandbox (never values).

Checking the app

ToolWhat it does
app_probeMake one HTTP request to the running app and report status, latency, content type, and a body snippet. The right way to confirm "is it actually serving?".
service_probeCheck that a bound service (database, etc.) is reachable from the sandbox.
deployment_statusRead the status of a sandbox's deployment(s) from the dashboard — status, URL, revision, message. Read-only; it never starts a deploy.

Python / data

ToolWhat it does
code_execRun Python in a persistent kernel — state carries across calls; charts render inline. Great for data work.
code_closeShut down a kernel and free its memory.

Built-in docs

ToolWhat it does
docs_readRead agentry's built-in cheat sheets baked into every sandbox — including the conventions for reading the signed-in user and framework setup tips.

Next

agentry — run AI-built apps on your own hardware.