IDE Setup

July 23, 2026 · View on GitHub

SuperLocalMemory V3 Documentation https://superlocalmemory.com | Part of Qualixar

Connect SuperLocalMemory to your AI coding tool. Once connected, memories are captured and recalled automatically.

Transport Options (v3.6.7+)

SLM ships with two MCP transports. Pick the one that fits your tool.

Transport	How it works	RAM cost	Requirement
HTTP (recommended)	All clients share one daemon process	~2 GB once, flat forever	`slm serve` must be running
stdio (universal)	One `slm mcp` subprocess per connection	~90–110 MB × connections	None — works offline
`mcp-remote` bridge	stdio wrapper that tunnels to HTTP	~50 MB bridge + HTTP pool	npm `@modelcontextprotocol/client-cli`

Quick rule: use HTTP if your tool supports it. Use mcp-remote if your tool is stdio-only but you want shared RAM. Fall back to pure stdio if you are offline or running daemon-free.

The mcp-remote bridge package:

npm install -g @modelcontextprotocol/client-cli

Auto-Detection

The fastest way to connect all your IDEs:

slm connect

This scans your system for installed IDEs, configures each one, and verifies the connection. Run it once after installing SLM.

To connect a specific IDE:

slm connect claude
slm connect cursor
slm connect vscode

Claude Code

Auto:

slm connect claude

Manual — HTTP (recommended, v3.6.7+):

claude mcp add --transport http superlocalmemory http://127.0.0.1:8765/mcp/

Or edit ~/.claude.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Manual — stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

Restart Claude Code. Verify with: slm status in a Claude Code session.

Claude Desktop

Claude Desktop uses a separate config file from Claude Code CLI.

Manual — HTTP (recommended, v3.6.7+):

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "superlocalmemory": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Manual — stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

Restart the Claude Desktop app after editing. Config is not hot-reloaded.

Cursor

Auto:

slm connect cursor

Manual — HTTP (recommended, v3.6.7+):

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Manual — stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "npx",
      "args": ["-y", "superlocalmemory", "mcp"],
      "env": {}
    }
  }
}

Restart Cursor. The memory tools appear in the tool list automatically.

VS Code / GitHub Copilot

Auto:

slm connect vscode

Manual — HTTP (recommended, v3.6.7+):

Add to VS Code settings (JSON):

{
  "mcp": {
    "servers": {
      "superlocalmemory": {
        "type": "http",
        "url": "http://127.0.0.1:8765/mcp/"
      }
    }
  }
}

Manual — stdio (fallback):

{
  "mcp": {
    "servers": {
      "superlocalmemory": {
        "command": "npx",
        "args": ["-y", "superlocalmemory", "mcp"]
      }
    }
  }
}

Windsurf

Auto:

slm connect windsurf

Manual — HTTP (recommended, v3.6.7+):

Add to ~/.windsurf/mcp.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Manual — stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "npx",
      "args": ["-y", "superlocalmemory", "mcp"],
      "env": {}
    }
  }
}

Gemini CLI

Auto:

slm connect gemini

Manual — HTTP (recommended, v3.6.7+):

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Manual — stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "npx",
      "args": ["-y", "superlocalmemory", "mcp"]
    }
  }
}

Antigravity IDE / agy CLI

Antigravity IDE and the agy CLI both read from ~/.gemini/antigravity/mcp_config.json.

HTTP (recommended, v3.6.7+):

{
  "mcpServers": {
    "superlocalmemory": {
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "npx",
      "args": ["-y", "superlocalmemory", "mcp"]
    }
  }
}

Restart Antigravity IDE after editing. The agy CLI picks up changes on next invocation.

Grok CLI

Grok CLI supports stdio only. Use the mcp-remote bridge to connect via HTTP:

npm install -g @modelcontextprotocol/client-cli

Add to ~/.grok/config.toml:

[mcp_servers.superlocalmemory]
command = "mcp-remote"
args = [
    "http://127.0.0.1:8765/mcp/",
    "--allow-http",
    "--transport",
    "http-only",
]

stdio (if mcp-remote is not available):

[mcp_servers.superlocalmemory]
command = "slm"
args = ["mcp"]

[mcp_servers.superlocalmemory.env]
SLM_AGENT_ID = "grok"

Hermes Agent

HTTP (recommended, v3.6.7+):

Edit ~/.hermes/config.yaml:

mcp_servers:
  superlocalmemory:
    type: http
    url: http://127.0.0.1:8765/mcp/

stdio (fallback):

mcp_servers:
  superlocalmemory:
    command: /opt/homebrew/bin/slm
    args:
      - mcp
    env:
      SLM_AGENT_ID: hermes

Both Hermes CLI and Hermes Desktop read the same ~/.hermes/config.yaml. Restart Hermes after editing.

Kimi

HTTP (recommended, v3.6.7+):

Add to ~/.kimi/mcp.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

CommandCode AICLI

HTTP (recommended, v3.6.7+):

Add to ~/.commandcode/mcp.json:

{
  "mcpServers": {
    "superlocalmemory": {
      "transport": "http",
      "enabled": true,
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

stdio (fallback):

{
  "mcpServers": {
    "superlocalmemory": {
      "transport": "stdio",
      "enabled": true,
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

JetBrains (IntelliJ, PyCharm, WebStorm, etc.)

Auto:

slm connect jetbrains

Manual: Open Settings > Tools > AI Assistant > MCP Servers and add:

Field	Value
Name	`superlocalmemory`
Command	`npx`
Arguments	`-y superlocalmemory mcp`

HTTP transport support varies by JetBrains version. Use stdio if HTTP is not available in your version.

Restart the IDE after adding the server.

Continue.dev

Auto:

slm connect continue

Manual: Add to ~/.continue/config.json:

{
  "mcpServers": [
    {
      "name": "superlocalmemory",
      "command": "npx",
      "args": ["-y", "superlocalmemory", "mcp"]
    }
  ]
}

Zed

Auto:

slm connect zed

Manual: Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "superlocalmemory": {
      "command": {
        "path": "npx",
        "args": ["-y", "superlocalmemory", "mcp"]
      }
    }
  }
}

Framework Adapters (v3.8.0)

Nine Python packages back their framework's native memory interface with the local SLM data root. Install alongside the framework; no running daemon is required for core usage.

Framework	Install name	Implements
LangGraph	`langgraph-superlocalmemory`	`BaseStore`
Semantic Kernel	`semantic-kernel-superlocalmemory`	`VectorStore`
Microsoft Agent Framework	`agent-framework-superlocalmemory`	`ContextProvider` / `HistoryProvider`
LangChain	`langchain-superlocalmemory`	`BaseChatMessageHistory`
LlamaIndex	`llama-index-storage-chat-store-superlocalmemory`	`BaseChatStore`
CrewAI	`crewai-superlocalmemory`	`StorageBackend`
AutoGen	`autogen-superlocalmemory`	`Memory`
Google ADK	`google-adk-superlocalmemory`	`BaseMemoryService`
OpenAI Agents	`openai-agents-superlocalmemory`	`SessionABC`

All adapters write to the same SLM data root that the CLI, MCP, and dashboard surface. Optional SLM providers, connectors, and backup have separate network behavior.

Full usage examples and prerequisites: Framework Adapters →

Verifying the Connection

After connecting any IDE, verify it works:

Open a chat/prompt session in your IDE
Ask: "What do you know about my preferences?"
If SLM is connected, the AI will check your memories before responding

Or run from the terminal:

slm status

Look for Connected IDEs: claude, cursor, ... in the output.

For HTTP transport specifically:

# Confirm the daemon is running and /mcp is live
curl -s http://127.0.0.1:8765/mcp/ \
  -X POST \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1"}}}' \
  | grep '"name":"SuperLocalMemory'

A response containing "name":"SuperLocalMemory V3" confirms HTTP MCP is live.

Troubleshooting

"slm command not found"

The npm global bin directory is not in your PATH.

# Find where npm installs global packages
npm root -g

# Add the bin directory to your PATH
# For zsh (~/.zshrc):
export PATH="$(npm root -g)/../bin:$PATH"

# For bash (~/.bashrc):
export PATH="$(npm root -g)/../bin:$PATH"

HTTP transport returns 404 or connection refused

The SLM daemon is not running. Start it:

slm serve
# or
slm restart

Then re-test with the curl command above.

IDE does not detect SLM tools

Ensure SLM is installed globally: npm list -g superlocalmemory
Restart the IDE completely (not just reload)
Check the MCP config file has correct JSON syntax
Run slm connect <ide> to regenerate the config

"Connection refused" or timeout errors

# Test the MCP server directly
npx superlocalmemory mcp --test

# Check for port conflicts
slm status --verbose

Multiple IDE configs conflicting

Each IDE has its own config file. They do not conflict. All IDEs share the same memory database at ~/.superlocalmemory/memory.db.

When using HTTP transport, all IDEs connect to the same daemon process at http://127.0.0.1:8765/mcp/. This is intentional — it is what keeps RAM flat.