Agent Tools (MCP) Asset

This documentation describes how to interact with the Celonis Agent Tools (MCP) Asset, which enables AI agents to securely access and interact with Celonis data and insights through the Model Context Protocol (MCP) and OpenAPI tool calling.

Goals

The Celonis Agent Tools (MCP) Asset provides a standardized way for AI agents to leverage Process Intelligence capabilities. It delivers a growing set of ready-to-use tools that AI agents can use to:

  • Search for relevant information from your Celonis data models and knowledge bases
  • Get real-time process context including KPIs, metrics, and process insights
  • Trigger actions in external systems based on process intelligence
  • Write back decisions to Celonis to create feedback loops and improve process execution

Delivered securely via the MCP, this toolkit makes it easy to build sophisticated agents that leverage Process Intelligence within any MCP-compatible client, such as Claude Desktop, Postman, or custom applications.

How it works

The Agent Tools (MCP) Asset is a Celonis asset type that encapsulates a collection of tools configured within the Celonis Platform. The asset supports both the MCP specification and OpenAPI tool calling, enabling AI agents to securely interact with external data sources and services.

Architecture

  1. Create and Configure : In Celonis Studio, you create an Agent Tools (MCP) Asset and configure it with the tools you want to expose to AI agents. These tools can include data retrieval functions, insight generators, knowledge model queries, and more.
  2. Publish : Once configured, the asset is published, generating a unique MCP Server URL that serves as the endpoint for MCP clients to connect.
  3. Authenticate : Clients authenticate using either OAuth 2.0 (recommended for production) or Application Keys (suitable for development and testing). The authenticated client receives a token that grants access to the Agent Tools (MCP) Asset.
  4. Discover Tools : Clients discover all available tools in the Agent Tools (MCP) Asset, including their names, descriptions, and input/output schemas.
  5. Execute Tools : AI agents can then call specific tools, passing the required parameters and receiving structured responses.

Rate Limiting

The API implements rate limiting to ensure fair usage and system stability:

  • Default rate limit: 500 calls per minute per asset