An operational mode in which an LLM or AI model is given a set of tools, allowed to determine appropriate tool use, adapt to a changing environment, and form synthesized conclusions based on tool output. This enables an LLM or AI model to autonomously decide its next actions.

In the context of MCP, agent mode allows an MCP Client to autonomously chain multiple tool calls across one or more MCP Servers to complete a complex multi-step workflow from a single prompt, without requiring the user to manually direct each action.

A sequence of actions autonomously planned and executed by an AI agent to complete a complex goal.

Agentic AI systems are independent AI agents that perform tasks by reasoning, learning, and adapting, and are radically changing how enterprises automate tasks and orchestrate complex workflows.

In MCP, agentic workflows are enabled by the MCP Client's ability to discover and invoke multiple tools across connected services in a coordinated sequence.

A software program that uses its own intelligence to complete instructed actions or achieve a desired outcome autonomously. Popular AI agents include OpenAI, Claude.ai, CrewAI, CoPilot, and Cursor, among others.

The process of verifying identity to ensure secure and authorized access between an MCP Client and an MCP Server. Zoho MCP supports two authorization types; Authorization on Demand and Authorization via Connection.

Authorization on Demand The default authorization method for Zoho MCP. It is account dependent, meaning each user connecting to the MCP Server authenticates and authorizes their actions individually using their own account credentials.

An optional organization-dependent authorization method where a Super Admin shares their OAuth Access and Refresh Tokens with trusted members of the organization, allowing them to authorize actions without individual authentication.

The information extracted from a user's natural language prompt that the MCP Client uses to determine which tools to invoke and what operations to perform in the connected service.

The environment or application that runs and manages one or more MCP Clients. The host is responsible for establishing and maintaining connections between the MCP Client and MCP Server.

The message serialization format used by MCP to encode and transmit messages between MCP Clients and MCP Servers. All MCP messages must be UTF-8 encoded JSON-RPC objects.

A type of AI model trained on vast amounts of text data, capable of understanding and generating human-like responses. LLMs form the core reasoning capability behind most AI agents and MCP Clients.

An open protocol developed by Anthropic that standardizes how AI agents connect to and interact with external tools, data sources, and services. It provides the required context to any AI agent to allow it to carry out complex workflows across different services.

The AI agent or application where a user provides a prompt to initiate a required workflow. MCP Clients such as Claude, OpenAI, Cursor, and Windsurf connect to MCP Servers to discover and invoke tools on behalf of the user.

A server that handles all backend operations required to allow MCP Clients to carry out actions in configured services. It manages MCP Client connections, processes the context provided through prompts, exposes tools to the MCP Client, and enforces lifecycle rules to ensure consistent state and access control.

A specific function or capability exposed by an MCP Server that an MCP Client can invoke to perform a defined action in a service. For example, in Zoho CRM, tools include GetRecords, SearchRecords, MassUpdateRecords, and UpdateNotes. The MCP Client can only perform actions defined by the tools that have been configured.

MCP URL or MCP Server URL is the access URL provided to an MCP Client to connect to and invoke a specific MCP Server and its configured tools.

An open standard authorization framework that allows secure delegated access to resources without sharing credentials directly. Zoho MCP uses OAuth tokens in its Authorization via Connection model.

A natural language instruction provided by a user to an MCP Client to initiate a desired workflow or action across connected services.

A pre-defined prompt structure provided by an MCP Server to guide the MCP Client in performing specific tasks within a service, helping ensure consistent and accurate tool invocations.

Data or content exposed by an MCP Server that an MCP Client can read or reference when carrying out operations in a connected service.

The target application in which the MCP Client performs actions via the MCP Server. For example, if you use an MCP Client to perform actions in Zoho CRM, then Zoho CRM is the service.

An earlier transport method used in MCP that allowed a server to push real-time updates to a client over HTTP.

The modern HTTP transport introduced in MCP to handle streaming interactions between clients and servers, released in early 2025 as an evolution of the earlier HTTP and SSE approach. In the Streamable HTTP transport, the server operates as an independent process that can handle multiple client connections using HTTP POST and HTTP GET requests, with the server optionally making use of SSE to stream multiple server messages. It simplifies how AI agents interact with services by using a single HTTP endpoint for sending and receiving responses between the client and server, replacing the need to implement separate endpoints for initializing the connection and sending messages.

The act of an MCP Client invoking a specific tool on an MCP Server to complete a task or retrieve information from a connected service.

The process by which an MCP Client automatically identifies and retrieves the list of available tools exposed by a connected MCP Server, allowing it to understand what actions it can perform without any manual configuration.

The communication method used to send and receive messages between an MCP Client and an MCP Server. The two standard transport mechanisms defined by MCP are stdio and Streamable HTTP.