> ## Documentation Index
> Fetch the complete documentation index at: https://docs.flowra.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Meta tools

> Discover, schema, execute, connect, and sandbox tools exposed in MCP Smart mode

Toolkit slug: **`flowra`**.

Meta tools are the **orchestration layer** for MCP Smart mode and advanced agents. They call other tools and platform APIs on your behalf. The hosted MCP URL binds exactly these six: four core tools and two sandbox tools.

| Slug | Name | What it does |
| - | - | - |
| `FLOWRA_DISCOVER_TOOLS` | Discover Tools | Search the catalog for tools matching a task; returns slugs and a suggested plan. Call first when the user mentions an app or action. |
| `FLOWRA_GET_TOOL_SCHEMAS` | Get Tool Schemas | Return full input/output JSON schemas for given tool slugs. Use before execute so arguments match the schema. |
| `FLOWRA_MULTI_EXECUTE_TOOL` | Multi Execute Tool | Run multiple action tools in one call (often after discover). |
| `FLOWRA_MANAGE_CONNECTIONS` | Manage Connections | Create or manage [connected accounts](/guides/connected-accounts); returns OAuth link URLs when needed. |
| `FLOWRA_JS_REMOTE_WORKBENCH` | Execute Code remotely in workbench | Run JavaScript in a persistent remote sandbox (state kept across calls). |
| `FLOWRA_REMOTE_BASH` | Remote Bash | Run shell commands in a remote sandbox; can load workbench JSON from multi-execute. |

## Workbench vs bash

Both sandbox tools share one sandbox per `session.id`, so they see the same files.

* **`FLOWRA_JS_REMOTE_WORKBENCH`** keeps JavaScript state between calls. It has helpers: `run_tool` (call a catalog tool), `invoke_llm`, `proxy_execute`, and `upload_local_file`. Use it to loop over many records or chain tool calls.
* **`FLOWRA_REMOTE_BASH`** runs one fresh shell per call with no helpers. Use it for `jq`, `grep`, `sort`, or `unzip` on files the workbench or multi-execute saved.

Neither sandbox can call `FLOWRA_MULTI_EXECUTE_TOOL` or another `FLOWRA_*` meta tool. Use `run_tool` inside the workbench for catalog tools.

## Recommended order

```mermaid theme={null}
sequenceDiagram
  participant AI as Agent / MCP client
  participant D as DISCOVER_TOOLS
  participant C as MANAGE_CONNECTIONS
  participant S as GET_TOOL_SCHEMAS
  participant M as MULTI_EXECUTE_TOOL

  AI->>D: User goal
  D-->>AI: tool slugs + plan
  alt Missing connection
    AI->>C: toolkit + auth
    C-->>AI: link URL or account id
  end
  AI->>S: slugs
  S-->>AI: schemas
  AI->>M: tool calls with args
```

## Related

* [MCP](/guides/mcp) — Smart mode exposes these tools.
* [Connected accounts](/guides/connected-accounts)
* [Native tools overview](/guides/native-tools)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.