MCP Server Documentation Overview
Understand the Model Context Protocol and how NeoArc Studio documents MCP servers. Learn about the three primitives (tools, resources, prompts), transport modes, authentication, and how MCP fits into your architecture documentation workflow.
The Model Context Protocol (MCP) is an open standard created by Anthropic that provides a uniform way for AI assistants to connect to external data sources and tools. Rather than building bespoke integrations for every system an AI agent needs to access, MCP defines a protocol layer that any server can implement and any client can consume.
NeoArc Studio treats MCP servers as a first-class integration type alongside REST APIs, GraphQL APIs, gRPC services, AsyncAPI event streams, and webhooks. Each MCP server is authored as a single .projection.json file with profileType: 'api' and apiType: 'mcp', capturing the server definition and each of its tools, resources, and prompts as projection entities.
The Three Primitives
MCP organises server capabilities into three complementary primitive types. Each serves a distinct role in how AI assistants interact with external systems.
How MCP Fits Into Your Architecture
MCP servers sit at the boundary between your internal systems and AI agents. They expose a controlled surface area that determines what the AI can see and do.
| Concern | MCP Solution | NeoArc Documentation |
|---|---|---|
| What can the AI do? | Tools with defined input/output schemas | Tool definitions with annotations, error scenarios, and examples |
| What data can the AI access? | Resources with URIs and MIME types | Resource definitions with template parameters and content previews |
| How should the AI behave? | LLM instructions and prompt templates | Server-level instructions and prompt definitions with message flows |
| What are the safety boundaries? | Tool annotations (read-only, destructive) | Semantic badges with colour-coded safety matrix |
| How does the AI connect? | Transports (stdio, HTTP) | Transport configuration with command, URL, and environment variables |
| Who is authorised? | OAuth 2.1, API keys | Authentication configuration with scopes and PKCE settings |
Storage
Each MCP server is a single .projection.json file with profileType: 'api' and apiType: 'mcp'. The projection can live anywhere in your workspace. Server-level configuration sits on apiProfile.protocolMetadata; tools, resources and prompts are projection entities within the same file.
{
"schemaVersion": "1.0.0",
"fileType": "projection",
"profileType": "api",
"apiProfile": {
"apiType": "mcp",
"protocolMetadata": {
"protocolVersion": "2025-06-18",
"capabilities": { },
"transports": [ ]
}
},
"entities": [
{ "entityKind": "mcp-tool", "name": "search-data" },
{ "entityKind": "mcp-tool", "name": "create-item" },
{ "entityKind": "mcp-resource", "name": "data-feed" },
{ "entityKind": "mcp-prompt", "name": "analyse" }
]
}Publishing and Viewing
When you publish your documentation site, MCP servers appear in the viewer sidebar as a navigable tree grouped by primitive type. Each server shows its tools, resources, and prompts with detail viewers that render annotations, schemas, examples, and transport configuration.
Next Steps
Create an MCP (Model Context Protocol) server projection with transports, capabilities and authentication. Document tools, resources and prompts with typed schemas.
Deep dive into MCP tool definitions: annotation flags, safety matrix, input/output schemas, error scenarios, result content types, related resources, and example patterns. Learn how to document tools that AI assistants can use safely and effectively.
Learn how to document MCP resources: concrete URIs, URI templates with parameters, content types, subscriptions, audience control, and metadata. Understand the difference between tools and resources and when to use each.