MCP Gateway

Overview

The MCP-based system has two trigger paths.

1. GitHub Issue Trigger
2. A GitHub Issue requests CT execution.
3. The Issue label selects the GitHub Self-hosted Runner or Jenkins automation path.
4. The shared Python Bridge calls mcp-server-local-runner or mcp-server-local-direct.
5. AI Agent Trigger
6. An AI Agent requests a tool through the VS Code MCP Gateway.
7. The Gateway discovers and routes tools to the Local MCP Server or GitHub MCP Server.

flowchart TD
    subgraph IssueTrigger["1. GitHub Issue Trigger"]
        direction TB
        Issue["GitHub Issue"]
        Automation["Windows Automation<br/>Self-hosted Runner or Jenkins"]
        RequestBridge["Request Bridge<br/>GitHub Issue Request to Local MCP Request"]
        ResultBridge["Result Bridge<br/>Local MCP Result to GitHub Issue Report"]
        IssueComment["GitHub Issue<br/>TEST Result Comment"]

        Issue --> Automation --> RequestBridge
    end

    subgraph AgentTrigger["2. AI Agent Trigger"]
        direction TB
        AgentRequest["AI Agent<br/>Tool Request"]
        RequestGateway["VS Code MCP Gateway<br/>Request Routing"]
        GitHubMCP["GitHub MCP Server"]
        ResultGateway["VS Code MCP Gateway<br/>Result Routing"]
        AgentResult["AI Agent<br/>Tool Result"]

        AgentRequest --> RequestGateway
        RequestGateway -->|GitHub request| GitHubMCP
        GitHubMCP -->|GitHub result| ResultGateway --> AgentResult
    end

    subgraph LocalMCP["Local MCP"]
        direction TB
        RunnerMCP["mcp-server-local-runner"]
        DirectMCP["mcp-server-local-direct"]
        SharedRuntime["Shared MCP Runtime<br/>runtime.py"]
        SharedTools["Shared TEST Tools<br/>Setup / Test / Log<br/>toolsets.py: extensible"]
        TestResult["MCP TEST Result<br/>JSON + Logs"]
    end

    RequestBridge -->|mode: runner| RunnerMCP
    RequestBridge -->|mode: direct| DirectMCP
    RequestGateway -->|Local MCP request| DirectMCP

    RunnerMCP --> SharedRuntime
    DirectMCP --> SharedRuntime
    SharedRuntime --> SharedTools --> TestResult

    TestResult -->|Issue execution result| ResultBridge --> IssueComment
    TestResult -->|AI Agent tool result| ResultGateway

    classDef sharedTools fill:#fff3bf,stroke:#f08c00,stroke-width:4px,color:#000;
    class SharedTools sharedTools;

Current MCP Source Mapping

The two Trigger paths are implemented with the following repository sources.

mcp/
  scripts/
    run_test_request.py       # Request Bridge
    make_test_result.py       # Result Bridge
  server_local_runner/
    server.py                 # Runner MCP entrypoint
  server_local_direct/
    server.py                 # Direct MCP entrypoint
  server_local/
    runtime.py                # Shared MCP runtime
    toolsets.py               # Shared TEST tools

Current source execution paths:

1. GitHub Issue Trigger: Self-hosted Runner

test_request_local.yaml
  -> run_test_request.py
  -> server_local_runner/server.py
  -> runtime.py + toolsets.py
  -> make_test_result.py
  -> GitHub Issue comment

1. GitHub Issue Trigger: Jenkins

Jenkinsfile
  -> run_test_request.py
  -> server_local_direct/server.py
  -> runtime.py + toolsets.py
  -> make_test_result.py
  -> GitHub Issue comment

1. AI Agent Trigger: Current Local MCP

.vscode/mcp.json
  -> server_local_direct/server.py
  -> runtime.py + toolsets.py
  -> MCP tool result
  -> AI Agent

MCP Purpose

• AI Agent to Local Tool connection
• AI Agent to GitHub resource connection
• common tool call path
• execution layer and collaboration layer split

AI Agent Connection Purpose

• shell command dependency reduction
• common interface for build, flash, test, log tools
• unified access to Issue, Pull Request, Review, Actions
• agent role based tool selection

Deployment Note

• Reference model
• Main AI, Sub AI, Local AI role split
• Practical model
• one Remote AI Agent can cover Main AI + Sub AI together
• Local AI is optional
• Interpretation
• role split is a documentation model
• runtime deployment can use 1, 2, or 3 agents

Role

Current Usage

1. VS Code reads .vscode/mcp.json.
2. VS Code connects MCP servers.
3. AI Agent selects tools by task.
4. Gateway routes the call to the matched MCP Server.

Notes:

• VS Code MCP Gateway is a connection hub.
• It is not a full workflow engine.
• end-to-end CT orchestration still needs workflow logic outside the gateway.

Routing Rules

Protocol Flow

sequenceDiagram
    participant A as AI Agent
    participant G as VS Code MCP Gateway
    participant L as Local MCP Server
    participant GH as GitHub MCP Server

    A->>G: tools/call { name, arguments }
    G->>G: match tool routing

    alt Local Tool
        G->>L: tools/call
        L-->>G: result
    else GitHub Tool
        G->>GH: tools/call
        GH-->>G: result
    else Unknown Tool
        G-->>A: tool_not_found
    end

    G-->>A: result

Gateway Scope

• covered
• MCP Server connection
• tool discovery
• tool call routing
• not covered
• GitHub event webhook receive
• result polling
• automatic result analysis chaining
• automatic final report posting

Agent Tool Access

Notes:

• This table is a role reference.
• One Remote AI Agent can handle both Main AI and Sub AI responsibilities.
• A single Remote AI Agent can use GitHub tools and result analysis tools together.
• Local AI is an optional helper, not a mandatory deployment unit.

CI/CD/CT Relation

Direct MCP Path

AI Agent
  -> VS Code MCP Gateway
  -> Local MCP Server or GitHub MCP Server

GitHub Issue Based CT Path

GitHub Issue
  -> GitHub Actions or Jenkins
  -> Python bridge
  -> Local MCP Server
  -> JSON / log / comment

Notes:

• test-request-runner
• GitHub Actions + self-hosted runner path
• test-request-direct
• Jenkins direct execution path
• both paths are outside the VS Code internal gateway scope

Related

• MCP Server-Local
• MCP Server-GitHub
• System Design