LM Studio

Configure Runstack MCP in LM Studio — Program → Install → Edit mcp.json with HTTP or stdio.

Source: https://runstack.engineer/docs/mcp-lmstudio

lmstudio

LM Studio (≥ 0.3.17 b10) acts as an MCP Host. Connect it to Runstack's MCP tools at https://app.runstack.engineer/mcp with Authorization: Bearer rnsk_…. Get a key at API Keys.

Add a server

Open LM Studio → Program tab in right sidebar → Install → Edit mcp.json.
Paste one of the configs below inside mcpServers and save.
LM Studio follows Cursor's mcp.json notation — supports both local and remote servers.

LM Studio also supports one-click deeplinks: Add to LM Studio buttons with base64-encoded config param. The manual JSON below is equivalent.

Remote (Streamable HTTP) — recommended

{
  "mcpServers": {
    "runstack": {
      "url": "https://app.runstack.engineer/mcp",
      "headers": {
        "Authorization": "Bearer rnsk_YOUR_API_KEY"
      }
    }
  }
}

Deeplink (optional)

https://lmstudio.ai/install-mcp?name=runstack&config=eyJ1cmwiOiJodHRwczovL2FwcC5ydW5zdGFjay5lbmdpbmVlci9tY3AiLCJoZWFkZXJzIjp7IkF1dGhvcml6YXRpb24iOiJCZWFyZXIgcm5za19ZT1VSX0FQSV9LRVkifX0=

Base64 decodes to the same url+headers above — replace placeholder before installing.

Local (stdio) — via `@rnsk/mcp`

{
  "mcpServers": {
    "runstack": {
      "command": "npx",
      "args": ["-y", "@rnsk/mcp"],
      "env": {
        "RUNSTACK_API_KEY": "rnsk_YOUR_API_KEY"
      }
    }
  }
}

When adding manually, copy only the object inside mcpServers — do not duplicate the outer key.

API variant (per-request ephemeral)

LM Studio also supports ephemeral MCP servers per API request (see Using MCP via API). For persistent use, the mcp.json config above is preferred.

Verify

Select a local model, open chat, and prompt: “Use Runstack to search for GitHub tools”. Check that search_tools appears in available tools and returns registry matches. Some MCP servers are heavy on tokens — if a local model overflows context, reduce tools via scoped prompts.

Gotchas

Never install untrusted MCP servers — they can run code and access files.
If tools do not appear, restart LM Studio after saving mcp.json and confirm JSON is valid (no trailing commas).