Build your first agent in just a few minutes with [Rasa Copilot](https://hello.rasa.com/?utm_source=docs&utm_medium=referral&utm_campaign=docs_cta).

**Rasa Tools are included in Rasa Pro 3.16 and later.** There is no separate package to install.

## What are Rasa MCP Tools?

Rasa MCP Tools are [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) tools that let agentic systems — IDE copilots, browser copilots, and other MCP clients — **inspect, validate, train, and debug** Rasa agents using natural language.

Building Rasa agents involves business logic spread across [flows](/content/docs/learn/concepts/calm/index.html), [slots](/content/docs/reference/primitives/slots/index.html), [custom actions](/content/docs/learn/guides/adding-custom-actions/index.html), and configuration files. IDE copilots are strong at coding but lack Rasa-specific knowledge. Rasa MCP Tools bridge that gap by making Rasa a **first-class tool provider** for your IDE.

### What you can do

| Capability | Example prompt |
| --- | --- |
| **Inspect** | "What flows and slots are in this project?" |
| **Build** | "Help me add a new subagent that manages flight bookings." |
| **Validate** | "Validate my project and explain any errors." |
| **Train** | "Train the model and tell me if it succeeded." |
| **Test** | "Send 'I want to transfer money' to the assistant and show me what happens." |
| **Evaluate** | "Generate and run scenarios for the add_contact flow." |
| **Debug** | "Get the assistant logs and explain why the flow was cancelled." |
| **Learn** | "Search the Rasa docs for how slot validation works." |

The MCP server exposes **19 tools** across six groups: documentation search, project introspection, schema retrieval, build/validation, runtime testing, and simulation & evaluation. For a complete tool-by-tool reference, see [Rasa MCP Tools API Reference](/content/docs/reference/api/rasa-mcp-tools/index.html).

### How it works

Your **IDE agent** starts `rasa tools run`, which is the **Rasa MCP server**: a local process that speaks MCP over stdio and reads your project files directly. You do not need a separate service for most tools. A **Rasa Server** started with `rasa run --inspect` is required for the runtime and evaluation tools: `talk_to_assistant`, `get_assistant_logs`, and `evaluate_agent`.

```text
Your machine

┌──────────────┐  MCP (stdio)   ┌──────────────────────┐
  │ IDE Agent    │ ◄────────────► │ Rasa MCP Server      │
  │              │                │ rasa tools run       │
  │ (your IDE)   │                │ reads project files  │
  └──────────────┘                └──────────┬───────────┘
                                             │ optional HTTP
                                             ▼
                                  ┌──────────────────────┐
                                  │ Rasa Server          │
                                  │ rasa run / inspect   │
                                  └──────────────────────┘
```

## Setup

### Prerequisites

- **Rasa Pro 3.16+** installed ( [Python](/content/docs/pro/installation/python/index.html) or [Docker](/content/docs/pro/installation/docker/index.html))
- **`RASA_LICENSE`** environment variable set. Some setups (for example **Claude Code**) start **non-interactive** shells, so they may not load `~/.zshrc`. On zsh, put the license where it is always read, such as `~/.zshenv`:

```bash
echo 'export RASA_LICENSE="your-license-key"' >> ~/.zshenv
```

### Quick setup with the wizard

Run from your **project root**:

```bash
rasa tools init
```

The wizard:

- Creates `.rasa/tools.yaml` with server configuration
- Downloads offline documentation into `.rasa/`
- Generates MCP configuration for your IDE
- Downloads Rasa specific skills for your project

Use `-y` to accept all defaults:

```bash
rasa tools init -y
```

After the wizard completes, open your IDE from the project root and check that `rasa-tools` appears in your IDE's MCP settings.

### Manual setup by client

If the wizard did not configure your IDE, or you prefer manual setup, follow the instructions for your client below.

### Cursor

1. Open **Settings > Tools & MCP** and add a new MCP server, or create `.cursor/mcp.json` in your project root.
2. Add the following config:

.cursor/mcp.json

```json
{
  "mcpServers": {
    "rasa-tools": {
      "command": "rasa",
      "args": ["tools", "run", "--mode", "stdio"]
    }
  }
}
```

3. Ensure `rasa` is on your `PATH`. Restart Cursor or reload MCP.
4. Open your project from the **project root**.

### VS Code (Agent Mode)

1. Create or open `.vscode/mcp.json` in your project root.
2. Add a server entry:

.vscode/mcp.json

```json
{
  "servers": {
    "rasa-tools": {
      "type": "stdio",
      "command": "rasa",
      "args": ["tools", "run", "--mode", "stdio"]
    }
  }
}
```

3. Reload the window or restart VS Code. The server will appear in the **MCP: List Servers** command palette entry.

### Claude Code

Claude Code launches non-interactive shells, so it may not inherit your terminal's `PATH`. Use the **full path** to `rasa`.

**Option A — CLI (recommended)**

From your **project root**:

```bash
claude mcp add rasa-tools -- /path/to/your/.venv/bin/rasa tools run --mode stdio
```

**Option B — edit `.mcp.json` directly**

Create or open `.mcp.json` in your project root:

.mcp.json

```json
{
  "mcpServers": {
    "rasa-tools": {
      "command": "/path/to/your/.venv/bin/rasa",
      "args": ["tools", "run", "--mode", "stdio"]
    }
  }
}
```

After either option, run `/mcp` in Claude Code and confirm `rasa-tools` is **Connected**.

### JetBrains (IntelliJ, PyCharm, WebStorm, etc.)

1. Open **Settings > Tools > AI Assistant > Model Context Protocol (MCP)**.
2. Click **Add** and choose **STDIO**.
3. Enter the following JSON config:

```json
{
  "mcpServers": {
    "rasa-tools": {
      "command": "rasa",
      "args": ["tools", "run", "--mode", "stdio"]
    }
  }
}
```

## Trying it out

After setup, open a copilot chat in your IDE and try these prompts:

```text
Use rasa-tools to summarize what this assistant can do.
```

```text
Help me use rasa-tools to create a new feature for this project.
Return:
1. User goal
2. Required slots (new vs reused)
3. Happy path
4. Edge cases
5. Test scenarios
Keep it under 15 bullets.
```

```text
Validate this Rasa project and explain any issues.
```

```text
Search the Rasa docs for how to add an mcp server to this assistant project.
```

If the tools respond with project information, your setup is working. See the [Prompt-Driven Agent Tutorial](/content/docs/learn/ai-assisted-development/index.html) for a guided walkthrough of building a feature with prompts.

For runtime testing (talking to the assistant) and simulation-based evaluation, you also need to start the Rasa server:

```bash
rasa run --inspect
```

Then you can test conversations:

```text
Talk to the assistant: "hello", "I want to send money to Jen", "$50", "yes".
Check if the transfer_money flow completed.
```

See [Simulation and Evaluation](/content/docs/reference/testing/evals/overview/index.html) for the full scenario YAML schema, assertion types, and result file format.

## Troubleshooting

| Issue | What to do |
| --- | --- |
| IDE says "no MCP servers found" | Check that the MCP config file exists in the right location and the JSON is valid. Reload or restart MCP in your IDE. |
| `rasa: command not found` | Ensure `rasa` is on your `PATH`. If using a virtualenv, use the full path to `rasa`, or point `command` at your `.venv/bin/python`. |
| `RASA_LICENSE` not found | Add `"env": { "RASA_LICENSE": "..." }` to your MCP server config. |
| Tools work but `talk_to_assistant` fails | Start the Rasa server first: `rasa run` or `rasa run --inspect`. |
| Offline docs are stale | Run `rasa tools init docs` and restart the MCP server. |
| First tool call is slow | Expected. The MCP server lazy-loads Rasa modules on first use. |
| Validation/training hangs | Check terminal for errors. |
| `evaluate_agent` fails to fetch tracker | The eval tool requires the Rasa REST API. Start your server with `rasa run --inspect`. |
| Stale MCP config | Restart the MCP server from your IDE and verify your tools are listed. |
| Skills not loading | Open your IDE from the **root directory of your assistant project**. |
| Skills list is empty | Run `rasa tools init` and select "Install Rasa agent skills" when prompted. |

For general Rasa Pro installation issues, see [Installation Troubleshooting](/content/docs/pro/installation/troubleshooting/index.html).