Quick Setup

Choose how you want to connect.

Use a hosted AI client such as Claude or ChatGPT, or connect a local model through LM Studio.

Path 1 — hosted AI

Claude or ChatGPT

Add the remote MCP server in your AI client, then use it in a new portfolio conversation.

Claude

  1. Open Customize → Connectors, then choose Add custom connector.
  2. Name it “MCP Portfolio Ideas” and add the hosted endpoint below.
  3. Add the API-key request header, save, then enable the connector in a new chat.
Claude’s connector instructions

ChatGPT

  1. Open Settings → Apps and create a custom MCP app/connector in Developer Mode.
  2. Name it “MCP Portfolio Ideas,” add the hosted endpoint, and choose the API-key authentication option.
  3. Scan the tools, create the app, then select it in a new chat before asking a portfolio question.
ChatGPT’s connector instructions
Remote MCP endpointhttps://mcp-portfolio-ideas.metapartner.ai/mcp/
Request headerX-API-Key: mcp_portf_ideas_guest_512B6vEMEMUZqJorLW9JlEzQrpMagnlwi-95QkqvxZk

Use the guest key only with non-sensitive examples. ChatGPT custom apps require an eligible plan or workspace with Developer Mode available.

Path 2 — local AI

Connect through LM Studio in three steps

Start with the shared guest workspace, verify the MCP tools are visible, then run a first portfolio analysis.

1

Copy the MCP config

Use the guest key for exploration. Anything sensitive should use a private user key instead.

2

Open LM Studio MCP settings

Paste the config into LM Studio > Developer > mcp.json, then reload tools.

3

Run a first analysis

Confirm the model can see the portfolio tools, then define a small portfolio and request metrics.

Open detailed setup guide

LM Studio configuration

Paste this JSON block into your MCP settings.

mcp.json 
"mcp-portfolio-ideas": {
  "url": "https://mcp-portfolio-ideas.metapartner.ai/mcp/sse/",
  "transport": "sse",
  "headers": {
    "X-API-Key": "mcp_portf_ideas_guest_512B6vEMEMUZqJorLW9JlEzQrpMagnlwi-95QkqvxZk"
  }
}
LM Studio MCP JSON configuration for MCP Portfolio Ideas
The guest workspace is shared and reset regularly. Use a private key for personal or sensitive portfolios.
Try This First

Example Prompts

Use these in order after the MCP tools show as available in your AI client.

See walkthrough
1

Discover available tools

Tell me what tools are available via the portfolio ideas interface.
2

Define the portfolio

I want to work on MyCurrentPortfolio, consisting of AAPL 25%, SPY 50%, COST 15%, SGOV 10%
2

Use a previously defined portfolio

What portfolios do I have available to work on? (primarily for registered users)
3

Analyze and explain

Analyze my current portfolio's performance. Explain whether the metrics indicate a good or poor performance.
3

Optimize Portfolio

On my portfolio "X", run an optimization to suggest a better allocation and explain the "why" behind it.
4

Add constraints

I want to add NVDA but keep my total stock exposure under 60%. Use the tools to find the best way to integrate it while staying in my risk profile.
4

Try different analyses

Optimize my portfolio for maximum returns.
4

Compare Performance

Compare and contrast optimized portfolios against each other.
MCP basics

Your AI assistant, connected to portfolio calculations

MCP is the connection that lets a compatible AI assistant use specialized portfolio-analysis tools instead of answering from general knowledge alone.

Frequently Asked Questions

Model Context Protocol (MCP) is an open standard that enables AI models like Claude or custom agents to securely interact with external data sources and tools. It provides a standardized way for the AI to "see" your portfolio and "use" our optimization tools.

Use the shared guest workspace only for exploration: its contents are shared and reset regularly. For personal portfolios, register for a private workspace and API key. We use authenticated access and encrypted transport for the service; see our Privacy Policy for what the service processes.

This is a hosted MCP server. Everything runs remotely, in the Google Cloud. No code is running on your machine, nothing is accessing your files, downloading or executing untrusted code. This is a plain communication plugin that sends requests to the MCP server, exactly the same way your LLM client, or the browser sends requests to Claude.AI or ChatGPT.com.
Review the access you grant to your AI client and use a private workspace before entering personal portfolio information.

My Personal preference, and heavy work horse, is LM Studio
The service is compatible with any AI assistant that supports the MCP standard, including Claude Desktop, various IDE extensions, and custom agentic frameworks.

Yes! You can use Streaming HTTP

Yes, a unique API key is provided in your configuration window above. This key allows your AI tool to authenticate its requests to our secure SSE endpoint. By default, you are using a shared "guest" key that enables you to answer the basic questions: how bad is my portfolio, and what configurations are better?". For a private workspace, private API key, and higher usage limits, register at Register a Private API Key.

It is a mathematical procedure that adjusts asset weights to either maximize returns for a specific risk level or minimize risk for a target return, based on historical correlations and volatility.

No. This service provides quantitative analysis and "ideas" for informative purposes. Execution and actual trading should always be handled through your choice of professional brokerage.

We currently offer a Free shared-guest tier at Builder level. Users who want a private workspace and higher usage limits can register for a free API key: Register a Private API Key. Enterprise accounts add advanced analysis and organization controls. See account capabilities and limits.

Tips

  • I use it with LMStudio: I use "openai/gpt-oss-20b" in high reasoning mode and only enable one tool that I use currently, mcp-portfolio-ideas to force the LLM to focus on my tools
  • When it gets confused, I reset the conversation: If you stop seeing the tools being called by the LLM, you might be getting answers directly from the LLM. As the LLM does not have access to the actual data and specialized portfolio calculations, it can start halluciating, which is highly undesirable. Always seek answers from the mcp-portfolio-ideas tools because they actually compute the answers using real market data.
  • Let the LLM interpret the results: After you get results, the LLM is more powerful interpreting what they mean, and what your next steps could be. Remember, mcp-portfolio-ideas is there to do the math and to consider the financial foundations, the decision is yours, so use LLMs, advisors, and your own judgement to make the best decisions.
    MCP-Portfolio-Ideas only guides you directionally, it does not provide solid investment advice
Warning

Using the MCP server configuration above puts you in the guest workspace, where everything is shared and visible. If you have portfolios or other information you consider private, don't use this workspace. Reach out to me for a private user key.
To facilitate experimentation, we're wiping clean the guest workspace every night at 4am EST