AI Agent Support

Connect an MCP-capable coding agent to official DevKit documentation and a local DevKit application with bdk mcp.

This integration is for local development. It does not expose a public HTTP endpoint or a production administration API.

What an agent can do

DevKit MCP gives an agent both implementation guidance and evidence from the running application. An agent can:

• search official DevKit documentation and API symbols
• get implementation guidance for common DevKit features
• inspect the selected runtime, registered modules, health, metrics, logs, and errors
• inspect messaging, queueing, jobs, and orchestrations when the host exposes them
• call project-owned diagnostics registered by the application
• run explicitly enabled operations after you approve them

The available runtime tools depend on the features registered by the selected application.

How the connection works

Your MCP client starts bdk mcp as a local process. The CLI finds a ready DevKit host in the current workspace and forwards runtime calls over local IPC.

Documentation and API search do not require a running host. Runtime diagnostics do.

Get started

MCP requires the bdk CLI, a DevKit web host with MCP enabled, and an MCP client configuration.

1. Install the bdk CLI

Install the CLI as a local .NET tool in the project repository:

dotnet new tool-manifest
dotnet tool install BridgingIT.DevKit.Cli

Confirm the installation:

dotnet tool run bdk --version

2. Enable MCP in the web host

Register MCP with the DevKit web application builder:

var builder = DevKitWebApplication.CreateBuilder(args)
    .AddConfiguration()
    .AddLogging()
    .AddModules(c => c
        .WithModule<CoreModule>())
    .AddMcp(c => c
        .WithHandlersFromAssembly<CoreModule>());

DevKit feature packages can register built-in handlers. Add project-owned handlers with .WithHandler<THandler>() or .WithHandlersFromAssembly<TMarker>().

3. Start the application

Run the DevKit web host in local development. The startup log includes a BDK line similar to:

[BDK] mcp handlers registered (...)

The host now has a local runtime descriptor and IPC endpoint.

4. Configure the MCP client

Github Copilot: For VS Code, add .vscode/mcp.json to the project repository:

{
  "servers": {
    "bdk": {
      "type": "stdio",
      "command": "dotnet",
      "args": ["tool", "run", "bdk", "mcp"]
    }
  }
}

Codex: For VS Code, add .codex/config.toml to the project repository:

[mcp_servers.bdk]
command = "dotnet"
args = [
  "tool", "run", "bdk", "mcp"
]
cwd = ".."
startup_timeout_sec = 60
tool_timeout_sec = 120
enabled = true

Other clients use different configuration files. Keep the command and arguments equivalent. See MCP Client Configuration for VS Code, Visual Studio, Rider, and other supported clients.

5. Verify the connection

Ask the agent:

Use the bdk MCP self-test and tell me whether the selected runtime is healthy.

The result should show one selected runtime in the ready state and list the operations advertised by the host.

Now ask the agent to use the documentation:

Use the bdk MCP docs tools to find the DevKit guidance for jobs. Summarize the recommended implementation pattern and list the relevant API types.

This request works without a running host because the CLI reads the official DevKit documentation and API index.

Go further

Use the AI Agent Support Guide for the architecture, tool catalog, coding workflow, prompt library, project-owned diagnostics, toolsets, and safety model.

Use the DevKit MCP Reference for the command and protocol contract.