AI Labnotes
Index
Developer infrastructure

Model Context Protocol

MCP architecture, servers, transports, security boundaries, integrations and debugging.

17mapped notes3 live · 3 scheduled · 11 research
Content graph / 17 nodes

Research path

Start with the pillar, move into a focused cluster, then follow its supporting field notes. Every node in the Model Context Protocol architecture is exposed here.

Live
3
Scheduled
3
Research
11
Parent–child pathPPillarCClusterSSupporting note
  1. 01PillarModel Context Protocol: Architecture, Servers, and RiskUnderstand MCP roles, primitives, transports and trust boundaries before choosing a server or writing an integration.model context protocolLive14,800 US/mo · KD 76
    1. 01.01ClusterMCP Servers: Choose, Verify, and ConnectChoose MCP servers by task fit, maintainer evidence, access scope and failure behavior—not by directory popularity.mcp serversLive18,100 US/mo · KD 70
      1. 01.01.01SupportingGitHub MCP Server: Setup, Permissions, and TestConnect GitHub's official MCP server with a narrow repository job, bounded credentials and observable read and write tests.github mcp serverLive6,600 US/mo · KD 62
      2. 01.01.02SupportingClaude MCP: Connect Servers to Claude CodeConfigure an MCP server in Claude Code at the correct scope, inspect it with /mcp and verify allowed and denied calls.claude mcpScheduled4,400 US/mo · KD 37
      3. 01.01.03SupportingContext7 MCP: Documentation Context Under TestInstall Context7 MCP for one documentation task, verify source identity and compare retrieved context with official docs.context7 mcpScheduled3,600 US/mo · KD 32
        1. 01.01.03.01SupportingJira MCP: Setup and Permission BoundariesConnect Jira through Atlassian Rovo MCP with bounded read, write and search permissions and direct record checks.jira mcpScheduled2,900 US/mo · KD 59
    2. 01.02ClusterBuild an MCP Server: A Reproducible PathBuild a minimal MCP server from one capability, a versioned contract and observable success and failure cases.build mcp serverResearchIntent mapped
      1. 01.02.01SupportingBuild an MCP Server in PythonBuild a Python MCP server with a pinned official SDK, narrow schemas and separate domain and protocol tests.build mcp server pythonResearchIntent mapped
      2. 01.02.02SupportingBuild an MCP Server in TypeScriptBuild a TypeScript MCP server around one typed capability, pin the SDK and verify schemas, errors and lifecycle.build mcp server typescriptResearchIntent mapped
      3. 01.02.03SupportingTest an MCP Server With MCP InspectorUse MCP Inspector to enumerate capabilities, send controlled payloads and localize server failures without overclaiming security.mcp inspectorResearchIntent mapped
    3. 01.03ClusterMCP Architecture: Host, Client, Server, and TransportMap MCP host, client, server and transport responsibilities so policy and protocol failures are debugged separately.mcp server architectureResearchIntent mapped
      1. 01.03.01SupportingMCP Client vs HostSeparate MCP host policy and user experience from each client's protocol session and server connection lifecycle.mcp client vs hostResearchIntent mapped
      2. 01.03.02SupportingMCP stdio vs Streamable HTTPChoose MCP stdio or Streamable HTTP by deployment boundary, identity, lifecycle and operations—not raw speed claims.mcp stdio vs streamable httpResearchIntent mapped
      3. 01.03.03SupportingMCP Tools, Resources, and PromptsChoose MCP tools, resources or prompts from user intent, control direction and side effects—not from naming preference.mcp tools resources promptsResearchIntent mapped
    4. 01.04ClusterMCP Security: Threat Model Before ConnectionThreat-model MCP data, tools, credentials and trust transitions before granting a server persistent access.mcp securityResearchIntent mapped
      1. 01.04.01SupportingMCP Authorization for Remote ServersImplement remote MCP authorization with OAuth requirements, resource-bound tokens, validation and no token passthrough.mcp authorizationResearchIntent mapped
      2. 01.04.02SupportingMCP Prompt Injection: Attack Paths and ControlsReduce MCP prompt-injection impact by treating tool and resource output as untrusted data and constraining consequential actions.mcp prompt injectionResearchIntent mapped