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).

On this page

Rasa MCP Tools are included in Rasa Pro 3.16 and later.

## v3.16

## Cheat Sheet

| Tool | Group | What it does |
| --- | --- | --- |
| `search_rasa_documentation` | Documentation | Search official Rasa docs |
| `list_project_flow_definitions` | Introspection | List all flows |
| `list_project_slot_definitions` | Introspection | List all slots |
| `list_project_response_definitions` | Introspection | List all responses |
| `get_flow` | Introspection | Get one flow by ID or name |
| `get_slot` | Introspection | Get one slot by name |
| `get_response` | Introspection | Get one response by name |
| `list_project_custom_actions_in_domain` | Introspection | List custom actions declared in domain |
| `list_custom_action_implementations` | Introspection | List custom action Python classes |
| `list_default_action_names` | Introspection | List built-in Rasa action names |
| `get_flow_schema` | Schemas | Get the flow JSON schema |
| `get_domain_schema` | Schemas | Get the domain YAML schema |
| `get_e2e_schema` | Schemas | Get the E2E test YAML schema |
| `validate_project` | Build | Validate project config and data |
| `train_rasa_assistant` | Build | Train the assistant model |
| `talk_to_assistant` | Runtime | Send messages to the running assistant |
| `get_assistant_logs` | Runtime | Get recent assistant logs |
| `validate_scenario` | Evaluation | Validate a scenario YAML before running |
| `evaluate_agent` | Evaluation | Run a simulated evaluation scenario |

## CLI Commands

| Command | Effect |
| --- | --- |
| `rasa tools init` | Interactive setup wizard. Creates `.rasa/tools.yaml` and downloads offline docs and skills. |
| `rasa tools init -y` | Non-interactive setup with all defaults. |
| `rasa tools init --project-path PATH` | Initialize for a project in a different directory. |
| `rasa tools init docs` | Download or refresh offline documentation files. |
| `rasa tools run --mode stdio` | Start the MCP server in stdio mode (for IDE integrations). |
| `rasa tools run --mode http --port 7331` | Start the MCP server in HTTP mode on the specified port. |
| `rasa tools run --config PATH` | Start the server using settings from `.rasa/tools.yaml` in the given directory. |
| `rasa tools run --project-path PATH` | Override the project folder the server reads from. |
| `rasa tools run --rasa-server-url URL` | Override the Rasa server URL (default: `http://localhost:5005`). |

## Documentation tools

### `search_rasa_documentation`
Search the official Rasa documentation for authoritative information. Returns relevant documentation about Rasa concepts, APIs, best practices, configuration, and troubleshooting with links to official docs.

| Parameter | Type | Required | Description |
| --- | --- | --- | --- |
| `query` | string | yes | The search query to find relevant Rasa documentation entries. |

**Returns:** Matching documentation snippets with source links.

**Sample prompts:**

```text
Search the Rasa docs for how to configure slot validation.
```

```text
Look up how collect steps work in Rasa flows.
```

### `list_project_flow_definitions`
List all flow definitions in the project. Returns flow ID, name, and file path for each flow.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `data_folder` | string | no | `"data"` | Folder containing flow YAML files, relative to project root. |

**Returns:** List of flows with `id`, `name`, and `file_path`.

**Sample prompt:**

```text
List all flows in this project.
```

### `list_project_slot_definitions`
List all slot definitions from the project domain file(s). Returns slot name, type, and file path.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `domain_folder` | string | no | `"domain"` | Folder containing domain YAML files, relative to project root. |

**Returns:** List of slots with `name`, `type`, and `file_path`.

**Sample prompt:**

```text
List all slots in this project and their types.
```

### `list_project_response_definitions`
List all response (utterance) definitions from the project domain file(s). Returns response name and file path.

**Returns:** List of responses with `name` and `file_path`.

**Sample prompt:**

```text
Show me all responses defined in this project.
```

### `get_flow`
Get a single flow by flow ID (YAML key) or flow name. Returns flow metadata and full definition including steps, triggers, and branching logic.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `flow_id` | string | yes | — | Flow ID (YAML key) or human-readable flow name. |
| `data_folder` | string | no | `"data"` | Folder containing flow YAML files. |

**Returns:** Full flow definition with metadata.

**Sample prompt:**

```text
Get the transfer_money flow and explain its steps.
```

### `get_slot`
Get a single slot by name. Returns slot metadata and full definition from the domain.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `slot_name` | string | yes | — | Slot name. |
| `domain_folder` | string | no | `"domain"` | Folder containing domain YAML files. |

**Returns:** Full slot definition with type, mappings, and metadata.

**Sample prompt:**

```text
Get the recipient slot definition and explain its mappings.
```

### `get_response`
Get a single response (utterance) by name. Returns response metadata and full definition including text, images, buttons, and custom payloads.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `response_name` | string | yes | — | Response name (e.g. `utter_greet`). |
| `domain_folder` | string | no | `"domain"` | Folder containing domain YAML files. |

**Returns:** Full response definition with all variations.

**Sample prompt:**

```text
Get the utter_ask_recipient response.
```

### `list_project_custom_actions_in_domain`
List all custom actions declared in the domain file(s). Returns action name and file path where the action is registered.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `domain_folder` | string | no | `"domain"` | Folder containing domain YAML files. |

**Returns:** List of custom action names with `name` and `file_path`. This lists domain declarations, not Python implementations.

**Sample prompt:**

```text
What custom actions are declared in the domain?
```

### `list_custom_action_implementations`
List all custom action Python implementations in the project. Returns action name, class name, and file path for each action.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `actions_folder` | string | no | auto-detected | Path to the actions folder relative to project root. Auto-detects from `endpoints.yml` or defaults to `"actions"`. |

**Returns:** List of action implementations with `action_name`, `class_name`, and `file_path`.

**Notes:**

- If `error` is set: the actions folder was not found. Check `endpoints.yml` or specify the folder.
- If `count=0` and no error: the actions folder exists but contains no action classes.

**Sample prompt:**

```text
List all custom action implementations and their file locations.
```

### `list_default_action_names`
List all built-in default action names provided by Rasa. These actions are available without configuration and can be overridden.

**Parameters:** None.

**Returns:** List of default action name strings.

**Sample prompt:**

```text
What are the default built-in Rasa actions?
```

## Schema tools

### `get_flow_schema`
Get the official Rasa flow schema in JSON Schema format. Use this to validate flow YAML or generate new flows.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `calm_only` | boolean | no | `true` | Return only CALM-related properties, excluding NLU-specific fields. |

**Returns:** JSON Schema document describing flow structure: name, description, step types, branching logic (`if`/`then`/`else`), collect steps, flow guards, and more.

**Sample prompt:**

```text
Get the flow schema so I can write a valid new flow.
```

### `get_domain_schema`
Get the official Rasa domain schema in YAML schema format. Use this to validate domain YAML or generate domain files.

**Returns:** YAML schema document describing domain structure: slots, custom actions, responses, and more.

**Sample prompt:**

```text
Get the domain schema. I need to add a new boolean slot correctly.
```

### `get_e2e_schema`
Get the official Rasa E2E test schema in YAML schema format. Use this to validate or generate end-to-end test files.

**Parameters:** None.

**Returns:** YAML schema document describing E2E test structure: test cases, steps (user and bot messages), fixtures, metadata, stub custom actions, and assertions.

**Sample prompt:**

```text
Get the E2E test schema, then write tests for the transfer_money flow.
```

## Build and validation tools

### `validate_project`
Validate the assistant project configuration and training data. Runs comprehensive checks on domain, flows, config, and training data.

**Parameters:** None (reads from the configured project folder).

**Returns:** Pass/fail status with a list of errors and warnings.

**Notes:**

- Can take 60+ seconds for large projects.
- Always run this after making changes and before training.

**Sample prompts:**

```text
Validate this project and tell me what's wrong.
```

```text
I made changes to the domain. Validate and fix any issues.
```

### `train_rasa_assistant`
Train the Rasa assistant with the current project configuration. Creates a new model in the `models/` directory.

**Parameters:** None (reads from the configured project folder).

**Returns:** Training status (success/failure), model path, and any errors.

**Notes:**

- Can take several minutes for large projects.
- Only call this after validation passes.
- Each training run produces a new timestamped model file.

**Sample prompts:**

```text
Train the assistant.
```

```text
Validate and then train. If validation fails, fix the issues first.
```

## Runtime testing and debugging tools

### `talk_to_assistant`
Test the assistant by sending a sequence of messages and verifying responses. Creates a new conversation for each call.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `messages` | list[string] | yes | — | List of user messages to send in sequence. Each message is sent after the assistant responds to the previous one. |
| `rasa_server_url` | string | no | `http://localhost:5005` | Override the Rasa server URL. Leave empty to use the configured default. |

**Returns:** Structured response with:

- Conversation history (user messages and assistant responses)
- Tracker context showing conversation state, active flows, and slot values

**Prerequisites:** The Rasa assistant must be running (`rasa run` or `rasa run --inspect`).

**Sample prompts:**

```text
Talk to the assistant: "I want to send money", "to Jen", "50 dollars", "yes".
Verify the transfer_money flow completes successfully.
```

```text
Test the happy path: send "hello" then "I need to transfer money to Bob".
Check which flow was triggered and what slots were filled.
```

### `get_assistant_logs`
Get recent log entries from the Rasa assistant for troubleshooting.

**Parameters:** None.

**Returns:** Recent log entries as text.

**Sample prompts:**

```text
Get the assistant logs and explain why the last conversation failed.
```

```text
Show me the assistant logs. I think a custom action is throwing an error.
```

## Simulation and evaluation tools

### `validate_scenario`
Validate a scenario YAML file before running any simulation. Checks syntax and structure, assertion type validity, and that all slot names referenced in `initial_slots`, `slot_was_set`, and `slot_was_not_set` assertions exist in the domain with compatible types. All errors are reported together in a single response.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `scenario_path` | string | yes | — | Relative path to the scenario YAML file within the project (e.g. `eval/scenarios/happy_path.yml`). |
| `domain_folder` | string | no | `"domain"` | Folder containing domain YAML files, relative to project root. Used to verify slot names and types. |

**Returns:** Pass/fail status with a list of all validation errors.

**Sample prompts:**

```text
Validate the scenario at eval/scenarios/order_delay.yml.
```

```text
I just edited a scenario file. Check it for errors before running.
```

### `evaluate_agent`
Run a simulation and evaluation loop for a given scenario. Loads `eval/conftest.yml`, simulates a multi-turn conversation with an LLM-based user against the running Rasa server, evaluates deterministic assertions against the tracker event history, and scores quality criteria and metrics with an LLM judge. Writes per-run result files and updates the experiment summary.

| Parameter | Type | Required | Default | Description |
| --- | --- | --- | --- | --- |
| `scenario_path` | string | yes | — | Relative path to the scenario YAML file within the project. |
| `experiment_timestamp` | string | yes | — | Experiment folder name (e.g. `2026-04-21_14-30-22`). Generate once per session with `datetime.now().strftime('%Y-%m-%d_%H-%M-%S')` and reuse across all scenarios so they land under the same folder. |
| `run_count` | integer | no | `1` | Number of times to run the scenario. Multiple runs surface LLM non-determinism. Capped at 10. |
| `rasa_server_url` | string | no | configured default | Override the Rasa server URL. Leave empty to use the server URL configured at startup. |

**Returns:** Pass/fail verdict, runs passed/total count, and the path to the updated `summary.txt`.

**Prerequisites:** The Rasa assistant must be running (`rasa run --inspect` recommended so that simulated conversations are viewable in the Inspector).

**Notes:**

- A run is marked failed if any deterministic assertion fails or any quality criterion fails.
- Quality metrics (`task_completion` and aggregated `bot_quality` — `helpfulness`, `repair_quality`, `coherence`, `tone`) are informational and do not gate the pass/fail verdict.
- On timeout, completed `run_N.txt` files are preserved but `summary.txt` is not updated.

**Sample prompts:**

```text
Generate and run a happy path scenario for the add_contact flow.
```

```text
Run the order_delay scenario 3 times and summarize the results.
```

```text
Run all scenarios in eval/scenarios/ and tell me which ones failed.
```

## Typical workflow

A common sequence for building a feature end-to-end:

1. **Discover** — `list_project_flow_definitions`, `list_project_slot_definitions`, `list_project_response_definitions`
2. **Understand schemas** — `get_flow_schema`, `get_domain_schema`, `get_e2e_schema`
3. **Implement** — write flows, domain entries, and custom actions
4. **Validate** — `validate_project`
5. **Train** — `train_rasa_assistant`
6. **Test** — `talk_to_assistant`
7. **Debug** — `get_assistant_logs` if behavior is unexpected
8. **Evaluate** — `validate_scenario`, `evaluate_agent` to run LLM-simulated conversations and score them against your goals