Rasa Model Context Protocol (MCP)
Product · Developer tool · Engineering & development by Rasa
A universal format for agents to access tools, services, and data.
Rasa Model Context Protocol (MCP) is a standardization tool for AI agents to access APIs and data. It enables developers to build modular, orchestrated systems with deterministic execution and state-preserving interruption handling.
Category: Developer Tools
- Protocol Standard
- Model Context Protocol (MCP)
- Integration Type
- API-based
- Execution Mode
- Deterministic or Inference-based
- Context Management
- Scoped memory and user-provided inputs
- Universal format for AI agents to access tools, services, and data
- Deterministic execution via declarative policies and rules
- State-preserving interruption handling for tool calls
- Modular tool definitions for reuse across flows and domains
Overview
Rasa Model Context Protocol (MCP) is a developer tool that provides a universal format for AI agents to access tools, services, and data. It is designed for businesses and developers building modular agentic systems that require clear execution control without hardcoding logic for every API. The protocol standardizes tool access, turning it into a discoverable and reusable resource that can be orchestrated across multiple systems. Rasa MCP enables deterministic execution by allowing tools to be triggered by rules or declarative policies rather than just model guesses. It also features built-in interruption handling, where tool calls pause if a user changes direction and resume without losing state once context stabilizes.
Standardization and Reuse
- Universal Tool Access: Standardize how AI agents connect with APIs, services, and data using a universal format.
- Modular Design: Define tool calling patterns once and reuse them across different agents, flows, and domains.
- Dynamic Discovery: Enable agents to access tools in real-time based on immediate needs without hardcoding logic.
Execution and Control
- Deterministic Execution: Trigger tools from declarative policies, conditions, or rules instead of relying solely on model inference.
- Interruption Handling Tracker: Pause tool calls during user interruptions and resume without losing conversation state once context stabilizes.
- Scoped Context Control: Explicitly define and scope the inputs and memory each tool receives to ensure focused execution.
Orchestration
- Orchestrated Access: Coordinate tool use across multiple systems with shared context and clear execution control.
- Autonomous Routing: Autonomously route tasks to the right agent, tool, or logic once roles and capabilities are defined.
Use cases
- Developers standardize how agents connect to internal APIs to ensure tools are reusable across different business domains.
- Architects implement deterministic tool triggering using rules to prevent models from making incorrect guesses during critical tasks.
- Teams build resilient conversational flows where tool calls pause and resume automatically when users interrupt with side questions.
- System integrators coordinate complex tool use across multiple legacy systems using a shared context and orchestration layer.
Specifications
- Protocol Standard
- Model Context Protocol (MCP)
- Integration Type
- API-based
- Execution Mode
- Deterministic or Inference-based
- Context Management
- Scoped memory and user-provided inputs
How it works
- Define Roles and Capabilities: Define the roles and capabilities for your agents and tools within the Rasa orchestration layer.
- Autonomous Routing: The orchestration engine autonomously routes tasks to the appropriate agent or tool based on the defined logic.
- Orchestrated Execution: Agents coordinate tool use across systems using shared context, with the ability to pause and resume during interruptions.
FAQ
Does Rasa MCP rely on model inference to trigger tools?
MCP allows developers to trigger tools from declarative policies, conditions, or rules, ensuring the model does not have to guess when to call a specific tool.
How does Rasa MCP handle user interruptions?
If a user changes direction mid-task, the MCP call pauses and the orchestrator tracks its state, allowing it to resume without losing progress once the context is stable.
Can I reuse tool definitions across different agents?
Yes, tool patterns are defined once and can be reused by agents across different flows and domains.
How is tool context managed in Rasa MCP?
The orchestration layer explicitly defines what each tool receives, from user-provided inputs to scoped memory, ensuring tools stay focused and secure.
Added by AntSpace from public sources (rasa.com).