Skip to content

GLOSSARY.md

Latest commit

 

History

History
104 lines (67 loc) · 2.83 KB

GLOSSARY.md

File metadata and controls

104 lines (67 loc) · 2.83 KB

Glossary

MCP and related terminology used in this project.

MCP Core Concepts

MCP (Model Context Protocol) : A protocol for AI models to access external tools, resources, and prompts in a standardized way.

MCP Server : A service that exposes tools, resources, and prompts via the MCP protocol. Built with Dedalus MCP.

MCP Client : An application (typically an AI assistant) that connects to MCP servers to use their capabilities.

Transport : The communication layer between client and server. Common transports: stdio, HTTP.

Primitives

Tool : A function the AI can invoke to perform actions. Has a name, description, and JSON schema for parameters.

Resource : Data the AI can read. Identified by URI. Can be static or dynamic.

Resource Template : A parameterized resource URI pattern (e.g., file://{path}). Allows dynamic resource generation.

Prompt : A reusable message template with optional parameters. Helps structure AI interactions.

Protocol Terms

Capability : A feature the server declares support for (tools, resources, prompts, etc.).

Initialize : The handshake between client and server. Exchanges capabilities and protocol version.

Notification : A one-way message that doesn't expect a response.

Request/Response : A message pair where the sender expects a reply.

Progress : Optional reporting of long-running operation status.

Server Concepts

Context : Request-scoped state available to tool/resource handlers. Access via get_context().

Decorator : The @tool, @resource, @prompt syntax for registering handlers.

Handler : A function that implements a tool, resource, or prompt.

Service : Internal component managing a category of handlers (ToolService, ResourceService, etc.).

Client Concepts

Connection : An active link to an MCP server. Manages lifecycle and communication.

Session : The period between initialize and disconnect. Maintains state.

Sampling : Client capability to generate AI completions. Server can request this.

Roots : Filesystem paths the client exposes to servers.

Authorization Terms

OAuth 2.1 : The authorization protocol used for secure MCP connections.

Bearer Token : A token included in requests to prove authorization.

Scope : Permissions granted to a token (e.g., tools:read, resources:write).

Framework Terms

MCPServer : The main class for building MCP servers in Dedalus MCP.

MCPClient : The main class for connecting to MCP servers.

Pydantic : The validation library used for type-safe parameters and responses.

See Also