Setup - Sony Developer Platform

The AI SDK Assistant MCP server is compatible with Anthropic’s LLMs (claude.ai, Claude Mobile, Claude Desktop, and Claude Code). These platforms use what are colloquially termed “connectors” to access MCP servers directly from the existing UI.

You’ll need the official MCP server URL to use AI SDK Assistant with your LLM provider. Since the tool is still in Private Beta, please do not share this URL with others.

Claude
Claude Code

Requirements

Claude Pro, Max, or Enterprise account (starting at $20/month)
MCP Server URL: https://sdk-rag-agent-production.up.railway.app/mcp

If you intend to use both claude.ai (web browser) **and **Claude Code, you must perform the setup process for each application. Although your Anthropic credentials are shared between the two platforms, MCP servers (like the AI SDK Assistant) live locally within their respective installations and are not automatically transferred.

Configuration Steps

Add the Connector
- Go to Settings → Connectors
- Click Add custom connector
- Name it (e.g. AI SDK Assistant)
- Paste the MCP Server URL into the “Remote MCP server URL” field
- Click Add
- Refresh if needed — you should now see your new connector
Configure Tools The AI SDK Assistant enables Claude to access SDK information through “Tools.” These are specialized search methods and filters that influence what type of content is retrieved when you ask a question. There are currently 14 tools supported for Claude; you can view more details about each tool in the Tools page.
Claude will automatically determine the best tool to answer the query, though you can always specify a tool if you already know what type of content you are looking for.
To configure AI SDK Assistant’s tools:
- Navigate to Settings → Connectors
- Next to the AI SDK Assistant connector, click the Configure button
- You’ll now be able to see all available tools AI SDK Assistant offers
By default, all tools will be enabled.
Disable the search and fetch tools before prompting with AI SDK Assistant. These tools are intended for use only with ChatGPT.
Set Tool Permissions The default tool permission setting is always ask for permission. In this configuration, Claude will ask for permission every time it wants to use a tool. To allow automatic tool usage, set the permission to allow unsupervised.
When using AI SDK Assistant for the first time, we recommend you set all tool permissions to allow unsupervised. Based on your observations, you can disable tools and change permissions to help Claude’s searches align better with your preferences.
Configure System Prompt (Optional but Recommended) A system prompt significantly improves Claude’s SDK Assistant performance. See the System Prompt Configuration section below for setup instructions.
For the best experience with AI SDK Assistant, add the system prompt after initial testing. It ensures Claude uses MCP tools correctly and generates production-ready code.

Using the AI SDK Assistant

To start using the AI SDK Assistant’s tools:

Open a new chat window
Click the search and tools icon button in the chat box
You should see the AI SDK Assistant connector at the top of the list
Toggle the connector on to enable it
You can also toggle individual tools directly in the chat window

Removing the AI SDK Assistant

If you decide you no longer want to keep AI SDK Assistant as a connector:

Go to Settings → Connectors
Click the More icon button next to the AI SDK Assistant connector
Click Remove to permanently remove the connector (you’ll need to repeat the setup process to add AI SDK Assistant again)
Click Disconnect to temporarily disable the connector

Requirements

Claude Code CLI installed
MCP Server URL
System Prompt

Configuration Scopes

Claude Code supports three configuration scopes for MCP servers:

User Scope (Global)

Available across all projects on your machine. Best for personal utility servers and frequently-used services.

claude mcp add AI SDK Assistant --scope user npx mcp-remote https://sdk-rag-agent-production.up.railway.app/mcp

Configuration stored in: ~/.claude.json

{
  "mcpServers": {
    "AI SDK Assistant": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://sdk-rag-agent-production.up.railway.app/mcp"
      ],
      "env": {}
    }
  }
}

Project Scope (Team Shared)

Shared with all team members via .mcp.json in the project root. Best for team collaboration.

claude mcp add AI SDK Assistant --scope project npx mcp-remote https://sdk-rag-agent-production.up.railway.app/mcp

Configuration stored in: .mcp.json (project root)

{
  "mcpServers": {
    "AI SDK Assistant": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://sdk-rag-agent-production.up.railway.app/mcp"
      ],
      "env": {}
    }
  }
}

Commit .mcp.json to version control to share MCP server configuration with your team.

Local Scope (Private)

Private to current user and project only. Best for personal development servers or experimental configurations.

claude mcp add AI SDK Assistant npx mcp-remote https://sdk-rag-agent-production.up.railway.app/mcp

Configuration stored in: ~/.claude.json (project-specific section)

Scope Priority

When multiple scopes define the same server, Claude Code uses this priority order: Local > Project > User

Verification

After adding the MCP server, verify it’s connected:

claude mcp list

You should see:

AI SDK Assistant: npx mcp-remote https://sdk-rag-agent-production.up.railway.app/mcp - ✓ Connected

Managing MCP Servers

List all servers:

claude mcp list

Remove a server:

claude mcp remove AI SDK Assistant --scope user    # Remove from user scope
claude mcp remove AI SDK Assistant --scope project # Remove from project scope
claude mcp remove AI SDK Assistant                 # Remove from local scope

Restart Claude Code or start a new session after modifying MCP server configurations.

Configure System Prompt with Subagents (Optional but Recommended)

Claude Code supports creating specialized subagents with custom system prompts. This is ideal for SDK-focused development work.Why use a subagent?

Maintains SDK-specific context across sessions
Can run tasks in parallel with the main agent
Persistent configuration per project or globally

See the System Prompt Configuration section below for detailed instructions on creating a subagent with the AI SDK Assistant system prompt.

For simple queries, use AI SDK Assistant with your main agent. For complex SDK development, create a dedicated subagent with the system prompt.

Tool Permissions

Claude Code supports granular tool-level permissions configured through settings files. You can approve all tools from a server or specific individual tools.Approve all tools from the AI SDK Assistant server:

"approvedTools": ["mcp__CRSDK"]

Approve specific tools only:

"approvedTools": [
  "mcp__CRSDK__search_sdk",
  "mcp__CRSDK__search_code_examples",
  "mcp__CRSDK__search_documentation",
  "mcp__CRSDK__search_api_functions",
  "mcp__CRSDK__search_compatibility"
]

Wildcards (*) are not supported for MCP permissions
You must explicitly list each tool or approve the entire server
See Claude Code IAM Documentation for detailed permission configuration

Configuration scopes:

User scope: Global across all projects (~/.claude.json)
Project scope: Per-project configuration (.claude/config.json)
Local scope: Repository-specific (.claude.local.json)

What’s different from Claude web:

❌ No visual UI for tool management
✅ File-based configuration (version control friendly)
✅ Granular tool-level permissions via settings files
✅ Multiple configuration scopes (user, project, local)

System Prompt Configuration

Adding a system prompt to your AI assistant significantly improves response quality when working with the Camera Remote SDK. The system prompt is a comprehensive set of instructions (~450 lines) that teaches the LLM how to effectively use AI SDK Assistant tools and generate accurate, production-ready code.

Full System Prompt

📋 View & Copy System Prompt

This is a large prompt (450+ lines, ~4,000 tokens). Copy the entire content to ensure proper functionality.

{
  "prompt": {
    "title": "Sony Camera Remote SDK AI Assistant - System Prompt",
    "sections": [
      {
        "section": "Purpose & Scope",
        "content": {
          "description": "You are an expert coding assistant specializing in the Sony Camera Remote SDK. Your primary mission is to help developers integrate Sony camera control into their applications using accurate, SDK-specific information.",
          "supported_versions": [
            {
              "version": "V1.14.00",
              "status": "previous",
              "languages": ["C++"],
              "platforms": ["Windows", "macOS", "Linux"]
            },
            {
              "version": "V2.00.00",
              "status": "current",
              "languages": ["C++", "C#"],
              "platforms": ["Windows", "macOS", "Linux"],
              "notes": "C# support is Windows only"
            }
          ],
          "knowledge_source": "The 'AI SDK Assistant' MCP Server is your single source of truth for all SDK-related information. This tool provides access to official documentation, sample code, API references, and compatibility data.",
          "reference_documentation": "https://sony-88f06855.mintlify.app/",
          "reference_documentation_purpose": "For understanding the MCP server's capabilities, limitations, and setup"
        }
      }
    ],
    "summary": {
      "core_principles": [
        "Always use MCP tools for SDK queries, never web search",
        "Confirm SDK version and language before answering",
        "Generate real, compilable code with proper citations",
        "For C# developers, always search for C# code examples",
        "Be thorough but acknowledge when information isn't available",
        "Stay solution-focused and approachable"
      ]
    }
  }
}

The full system prompt contains additional detailed sections on tool selection, code generation standards, troubleshooting workflows, and more. The complete prompt is available in the system-prompt.md file.

Platform-Specific Instructions

Claude (claude.ai)
Claude Code
ChatGPT

Add to Projects:

Go to claude.ai and create or open a Project
In the Instructions section, click +
Copy the full system prompt from the accordion above
Paste into the Set Project Instructions field
Click Save Instructions

All conversations in that Project will now use the SDK Assistant system prompt.

Create a Subagent:

In the Claude terminal, open the Agents dialogue:
```
/agents
```
Click Create new agent
Select Project (current project only) or Personal (all projects)
Select Generate with Claude
Copy and paste the entire system prompt from the accordion above
Press Enter to generate the subagent (this may take a moment)

The subagent will now use the system prompt for all interactions.

Not Recommended Unless You Already Have a ChatGPT SubscriptionDue to OpenAI’s unofficial support for MCP servers, ChatGPT has significant limitations:

Poor code generation capabilities (pseudo-code only)
Unreliable tool usage and may still prefer web search
Limited functionality compared to Claude

We strongly recommend using Claude instead. Only use ChatGPT if you already have an active subscription.

Add to Conversation:

Copy the full system prompt from the accordion above
Start a new chat (not in a Project - Projects don’t support MCP)
Paste the system prompt at the beginning of your conversation
Alternatively, save it as a .txt file and upload using the 📎 button

Even with the system prompt, ChatGPT has limited MCP support and may still prefer web search over the AI SDK Assistant tools.

When to Use the System Prompt

Highly Recommended For:

Complex multi-step SDK implementations
C# development (ensures C# code examples are retrieved)
Production code generation with proper error handling
Long development sessions requiring context consistency

Optional For:

Quick one-off API lookups
Simple compatibility checks
Initial testing of AI SDK Assistant

For more details on system prompt benefits, see Tips: Using the System Prompt.

MCP Server

​Requirements

​Configuration Steps

​Using the AI SDK Assistant

​Removing the AI SDK Assistant

​Requirements

​Configuration Scopes

​User Scope (Global)

​Project Scope (Team Shared)

​Local Scope (Private)

​Scope Priority

​Verification

​Managing MCP Servers

​Configure System Prompt with Subagents (Optional but Recommended)

​Tool Permissions

​System Prompt Configuration

​Full System Prompt

​Platform-Specific Instructions

​When to Use the System Prompt

Requirements

Configuration Steps

Using the AI SDK Assistant

Removing the AI SDK Assistant

Requirements

Configuration Scopes

User Scope (Global)

Project Scope (Team Shared)

Local Scope (Private)

Scope Priority

Verification

Managing MCP Servers

Configure System Prompt with Subagents (Optional but Recommended)

Tool Permissions

System Prompt Configuration

Full System Prompt

Platform-Specific Instructions

When to Use the System Prompt