Skip to content

INSTALL-PLATFORMS.md

Latest commit

 

History

History
209 lines (136 loc) · 6.57 KB

INSTALL-PLATFORMS.md

File metadata and controls

209 lines (136 loc) · 6.57 KB
Raw

MARM v2.12.1 MCP Server - Platform Integration Guide

Table of Contents

Overview: Connecting MARM to Apps & Platforms

This guide provides platform-specific instructions for integrating the MARM MCP Server with major AI applications and developer platforms.

Important: This covers base applications and API integrations - not CLI tools. For CLI setup, see the dedicated docs folder for install guides Docs

Part 1: Base Application Integration

Claude (Anthropic)

Claude Web App (claude.ai)

Supported: ✅ Remote MCP servers Requirements: Pro, Max, Team, or Enterprise plans

Setup Process:

  1. Log in to claude.ai in your browser
  2. Go to Settings > Connectors
  3. Click "+ Add custom connector"
  4. Enter your MARM server URL: http://your-server.com:8001/mcp
  5. Tools automatically available in web and mobile apps

Claude Mobile Apps (iOS/Android)

Supported: ✅ Remote servers only (no localhost) Requirements: Pro+ plans, must configure via web first

Setup Process:

  1. First configure via claude.ai web app (above)
  2. Mobile app automatically syncs connected tools
  3. Enable/disable tools per chat in mobile app settings

Claude Workbench (Developer)

Supported: ✅ Full MCP integration Requirements: Anthropic API access

Setup Process:

  1. In Workbench interface, select "Tools" or "MCP Server" integration
  2. Add MARM URL: http://localhost:8001/mcp
  3. Define tool usage in your agent configuration
  4. Test tool calls with prompt triggers

ChatGPT (OpenAI)

ChatGPT Base App (chat.openai.com)

Supported: ✅ Developer Mode only (Sept 2025 update) Requirements: ChatGPT Pro, Business, Enterprise, or Education plans

Setup Process:

  1. Enable Developer Mode in ChatGPT settings
  2. Navigate to Connectors section
  3. Add custom connector with MARM URL: http://your-server.com:8001/mcp
  4. Supports both read and write actions

Limitations:

  • Remote servers only (no localhost support)
  • Implementation is basic compared to Claude
  • GPT-5 not fully MCP-ready yet

OpenAI Playground

Supported: ✅ Testing only (does not execute) Requirements: OpenAI API access

How it works:

  • Playground generates tool calls but doesn't execute them
  • Shows function call output for inspection
  • Manually provide tool responses to test model behavior
  • Prevents accidental real-world actions during development

Gemini (Google)

Gemini Base App (gemini.google.com)

Supported: ❌ No MCP support

Google AI Studio

Supported: ⚠️ Limited function calling (not full MCP) Requirements: Google AI Studio access

How it works:

  • Generates function calls as structured output
  • Does not execute tools during testing
  • Must use Gemini SDK in your own application for execution
  • Manual tool definition required (no auto-discovery)

Grok (xAI)

Grok Base App (grok.com & Mobile)

Supported: ❌ Not compatible with standard MARM deployment

Grok's MCP connector support requires a public HTTPS URL. MARM runs on localhost:8001 by default and Docker binds to 127.0.0.1 — xAI's servers cannot reach either. You would need to host MARM on a public server with a domain and TLS certificate for this to work.

Grok Developer Platform

Supported: ✅ Remote MCP via xAI API Requirements: xAI Developer API access

Current Status:

  • Remote MCP Tools support Streaming HTTP and SSE transports
  • localhost does not work — xAI connects from its own infrastructure
  • Requires a publicly resolvable HTTPS MARM endpoint
  • Authenticate by passing authorization: "Bearer <MARM_API_KEY>" in the tool config header

Part 2: Developer Integration

Adding MARM to Custom Applications

If you're building your own application with any LLM provider, here's the basic pattern:

1. Install MCP Client:

pip install mcp

2. Connect to MARM:

from mcp.client.streamable_http import streamablehttp_client
from mcp import ClientSession

async with streamablehttp_client("http://localhost:8001/mcp") as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()
        result = await session.call_tool("marm_smart_recall", {"query": "test"})

3. Use with your LLM:

  • OpenAI: Pass tools to chat.completions.create()
  • Anthropic: Use tools parameter in Claude API
  • Google: Convert to Gemini's function calling format
  • Grok: Use xAI Remote MCP Tools with your hosted MARM endpoint

Each provider has different API patterns, but the MARM connection stays the same. If you're developing, you already know how to integrate tools with your chosen LLM platform.

Platform Comparison Summary

Platform Base App Support Developer Support MCP Ready Notes
Claude ✅ Full (Remote) ✅ Full ✅ Yes Best MCP support
ChatGPT ⚠️ Developer Mode ✅ API + Playground ⚠️ Basic Sept 2025 addition
Gemini ❌ None ⚠️ Function Calling ❌ No Manual definitions required
Grok ❌ None (requires public HTTPS) ✅ API Remote MCP ✅ Yes Requires public HTTPS endpoint; no localhost

Best Practices

For Production Applications:

  • Use remote MARM deployment (not localhost)
  • Implement proper error handling for tool calls
  • Cache frequently accessed memories
  • Monitor tool usage and performance

For Development:

  • Start with Claude for best MCP experience
  • Use Playground/Studio for testing without execution
  • Test with multiple providers for compatibility
  • Implement fallback for unsupported platforms

Need Help? Check our README.md or join Discord for integration support.