Skip to content

Client Setup

Connect your AI client to the JickleJime MCP server to give it access to your campaign knowledge base.

The repository includes an opencode.json project config that registers the local jicklejime MCP server:

{
  "mcp": {
    "jicklejime": {
      "type": "local",
      "command": [
        "dotnet",
        "run",
        "--project",
        "src/JickleJime.Mcp",
        "--",
        "--stdio"
      ],
      "environment": {
        "DOTNET_ENVIRONMENT": "Development"
      },
      "enabled": true
    }
  }
}

Start opencode from the repository root. The jicklejime MCP tools are loaded from the project config and are available to the built-in build/plan agents and the repo-local custom agents.

To verify the server is registered, run:

opencode mcp list

If the first tool load times out after a clean build or SDK restore, restart opencode and try again. The project config gives the MCP server a longer startup timeout than opencode's default.

Claude Desktop

Add the server to your Claude Desktop configuration file:

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "jicklejime": {
      "command": "dotnet",
      "args": [
        "run",
        "--project", "/path/to/jicklejime/src/JickleJime.Mcp",
        "--", "--stdio"
      ],
      "env": {
        "DOTNET_ENVIRONMENT": "Development"
      }
    }
  }
}

Replace /path/to/jicklejime with the actual path to your clone of the repository.

Other local clients

Other MCP-compatible clients can still connect by launching the server in stdio mode:

DOTNET_ENVIRONMENT=Development dotnet run --project src/JickleJime.Mcp -- --stdio

Client-specific configuration shapes differ. Use the command above as the shared launch contract.

Remote / deployed server

For a deployed MCP server accessible over HTTP:

  1. Endpoint — https://your-server/mcp
  2. Transport — HTTP Streamable
  3. Authentication — Authorization: Bearer <api-key> (if an API key is configured)

Configure your client to connect to the HTTP endpoint instead of launching a stdio subprocess. The exact configuration depends on your client.

Prerequisites

Before the MCP server can connect to your knowledge base, make sure:

  1. PostgreSQL is running — docker compose up -d starts the local database
  2. Credentials are configured — Azure OpenAI endpoint, PostgreSQL connection string, and optionally Azure AI Document Intelligence (see Configuration Reference)
  3. .NET 10 SDK is installed — Required to build and run the server
  4. Documents are ingested — Use the CLI to ingest documents before searching

Troubleshooting

"No results found" — Make sure documents have been ingested. Run jime list to check.

Connection refused — Verify the database is running (docker compose up -d) and credentials are configured.

Authentication errors — If using the HTTP transport with an API key, ensure the Authorization: Bearer <key> header is included in requests.