MCP Glossary

.mcpb Extension File A zip-compressed open-standard packaging format for MCP servers.

Actions The result of an AI agent successfully calling a tool to perform work. Agentic AI Intelligent programs that autonomously pursue goals and take actions. AI Interoperability Ability of AI systems and data sources to work together seamlessly. Air-Gapped MCP A high-security deployment pattern where MCP servers operate on a network completely isolated from the public internet. Annotations Optional metadata fields to assist AI models in handling data. API A set of rules for software interaction, standardized by MCP for AI. Assistant Message A message generated by the AI model in response to a user prompt or tool output. Atomic Operations Protocol operations that are guaranteed to either fully succeed or fully fail, leaving no partial or inconsistent state. Authentication Verifying a client's identity before allowing tool or resource access. Automatic Tool Calling A mode where the AI application automatically determines when and how to call MCP tools without needing a specific prompt for each step.

Bidirectional Connections Two-way data flow between AI models and external tools. Bundle A cloud-hosted collection of related MCP tools under a single URL.

Caching (MCP) The storage of previous MCP response data to speed up subsequent requests and reduce token consumption. Capabilities Declared features that an MCP client or server support. Capability Negotiation Agreement on supported features during the handshake phase. Client Info Metadata provided by an MCP client during the handshake to identify itself and its version. Client The AI application or agent connecting to an MCP server. Command Line Arguments Values passed to an MCP server at startup to configure its behavior and connection parameters. Communication Protocol The JSON-RPC 2.0 based format for client-server messages. Context The surrounding information that informs an AI model's response. Context Window The working memory or total processing capacity of an LLM. Context Window Management The active management of available tokens in an AI's memory to ensure the most relevant information is preserved. Contextual Metadata Tagging Attaching descriptors to resources for better model understanding. Create-MCP-Server A CLI utility for scaffolding new MCP server projects. Cursor An opaque string used in pagination to represent a pointer to the next set of results.

Data Sovereignty The right of an individual or organization to maintain control over the storage and use of their data within MCP systems. Docker MCP A deployment pattern where MCP servers are packaged and executed inside Docker containers. Dynamic Schema MCP tools or resources whose structure and available options change based on runtime state or specific user context.

Environment Variables Dynamic-named values that can affect the way an MCP server behaves, often used for secrets and API keys. Error Handling Standardized JSON-RPC error codes and messages in MCP.

Handshake The initial negotiation phase for client-server connection. Host The application or environment where an MCP client is running.

Idempotency The property of an MCP operation where performing it multiple times with the same parameters has the same effect as performing it once. Inspector An interactive developer tool for testing MCP servers. Introspection The process of discovering tools and capabilities from a server.

JSON-RPC 2.0 The underlying text-based protocol for all MCP messages. JSON Schema Defining the structure of input arguments for MCP tools. JMESPath Pruning A high-performance JSON query language used to optimize LLM context.

Lifecycle The stages of an MCP connection from init to shutdown. Local Server An MCP server that runs on the user's local machine via stdio. Logging Real-time status and error messages from server to client.

Max Tokens The upper limit on the number of tokens an AI model can generate in a single response. Model Context Protocol (MCP) An open standard for universal AI-to-system integration. MCP Audit Log A searchable, immutable record of every interaction between an MCP client and server for security and compliance monitoring. MCP Best Practices Recommended architectural and implementation patterns to ensure high performance and reliability in MCP systems. MCP Bridge A component that translates between the Model Context Protocol and other communication standards or external APIs. MCP Compliance The adherence of an MCP implementation to security, privacy, and regulatory standards. MCP Composition Unified API mesh bridging multiple MCP servers. MCP Credential Rotation The automated process of periodically changing the secrets (API keys, tokens) used by MCP servers to access external services. MCP Health Check A period check performed by an MCP client to ensure a server is still running and responsive. MCP Hub A centralized platform or repository for discovering, sharing, and managing MCP servers. MCP Icon A visual representation of an MCP server, typically provided as a URL or a base64-encoded string. MCP Manifest A configuration file that describes a server's identity, entry points, and required permissions. MCP Metadata Additional descriptive data attached to MCP resources, tools, or prompts to provide specialized context to clients. MCP Middleware A layer that sits between an MCP client and server to intercept, modify, or audit protocol messages. MCP on Lambda Running MCP servers as serverless functions on AWS Lambda for infinite scalability and zero-maintenance infrastructure. MCP over VPN Connecting a local MCP client to a remote MCP server through a Virtual Private Network for secure, encrypted communication. MCP Performance Tuning Optimizing MCP server performance to reduce latency, token usage, and resource consumption. MCP Telemetry Real-time observability and usage metrics for the agentic layer. MCP Tracing The capturing and analysis of the path of a request through an MCP system to debug performance and reliability. Model Preferences Hints or requirements provided to an AI host regarding which specific model or model family should be used for a task. Multi-Server Orchestration The management and coordination of multiple MCP servers by a single host application to solve complex, cross-domain tasks.

OAuth2 Flow Dynamic authentication elicitation for agent access. OpenAPI-to-MCP The automated creation of an MCP server based on an OpenAPI (Swagger) specification.

Pagination The process of dividing a large set of resources or tool results into smaller, manageable chunks called pages. PII Masking Privacy feature that redacts sensitive info from API responses. Prompts Pre-defined templates offered by servers to guide models. Prompt Injection Protection Security measures designed to prevent malicious actors from subverting an AI's instructions through carefully crafted inputs to MCP servers. Protocol Extensions Custom additions to the core Model Context Protocol that allow for specialized data types or communication patterns. Protocol Versioning Negotiating a common protocol version for compatibility. Provider The underlying external API service (e.g., Salesforce, Slack).

Rate Limiting A strategy used by MCP hosts or hubs to control the frequency of requests sent to a server. Recursive MCP An advanced pattern where an MCP server acts as a client to another MCP server to solve complex tasks. Resource Notifications Proactive updates when a resource's content changes. Resource Template A URI pattern provided by an MCP server that allows clients to resolve dynamic resources with specific parameters. Resource URI A unique address identifying a specific piece of data. Resources Static data objects that provide context to an agent. Roots Base directories shared by a client with an MCP server.

Sampling Server-initiated requests for model text generation. Schema Generation The automated creation of JSON schemas from code or data models to define MCP tool arguments and resource types. SDKs Software development kits for easily building MCP servers. Server Info Details provided by an MCP server during the handshake including its name and version. Server ID Unique identifier for a server configuration in a client. Server The bridge providing tools and context to AI clients. SSE HTTP-based streaming protocol for remote MCP servers. Stdio The standard OS streams used for local MCP transport. Stop Sequences Instructions provided to an AI model during sampling to halt generation when a specific string is encountered. System Prompt The foundational instructions provided to an AI model to define its persona, constraints, and operational guidelines.

Temperature A parameter that controls the randomness of an AI model's output during sampling. Tool Choice A control parameter that allows a user or developer to force the AI to use a specific tool, no tool, or choose automatically. Tool Naming Conventions Patterns like action_noun for clear tool selection. Tool Notifications Real-time updates when a server's toolset changes. Tool Selection The AI model's process of choosing the best tool to call. Tools Granular, callable operations provided by an MCP server. Transport The communication layer carrying bytes (e.g., SSE, stdio).

URI Templates Flexible URI patterns with placeholders for dynamic data. User Approval Security model requiring human consent for sensitive actions. User-in-the-Loop A critical safety pattern where the AI agent must obtain explicit human permission before executing high-risk tool calls. User Message A message in an AI conversation that represents input or instructions from the human user.

WebAssembly (Wasm) Servers Sandboxed, portable deployment pattern for MCP servers. WebSocket Transport A full-duplex communication protocol for real-time, bidirectional MCP interactions over a single TCP connection.