MCP server for LM Studio: add Sume's hosted MCP

To add an MCP server to LM Studio, use version 0.3.17 or later, open the Program tab in the right sidebar, click Install > Edit mcp.json, and add the server under mcpServers. A remote server needs a url and, when it asks for a token, a headers entry. Sume's hosted MCP server fits that shape: https://mcp.sume.com/mcp with your Sume API key in an Authorization: Bearer header gives your local model image, video, and audio tools that run on Sume's servers.

LM Studio's side comes from its MCP docs and its 0.3.17 release notes; Sume's side comes from MCP OAuth and API keys, MCP tools and gates, and Jobs and results, all read on 2026-09-27. Sume has no official LM Studio integration: LM Studio connects to Sume's remote MCP server like any other, and Sume's basics page says hosted MCP still works but is not part of the primary path today.

How do I add Sume to LM Studio's mcp.json?

LM Studio acts as an MCP host from 0.3.17, supports local and remote servers, and follows Cursor's mcp.json notation. The file lives at ~/.lmstudio/mcp.json on macOS and Linux and at %USERPROFILE%/.lmstudio/mcp.json on Windows, but LM Studio recommends its in-app editor. When you save the file, LM Studio loads the servers in it. Replace <SUME_API_KEY> with a key from Sume's dashboard:

{
  "mcpServers": {
    "sume": {
      "url": "https://mcp.sume.com/mcp",
      "headers": {
        "Authorization": "Bearer <SUME_API_KEY>"
      }
    }
  }
}

How does LM Studio authenticate to an MCP server?

With a header, as far as LM Studio's docs go: its MCP page shows a remote server authenticated by an Authorization: Bearer token in headers and describes no OAuth sign-in. That matches Sume's API-key path: hosted MCP accepts Authorization: Bearer <SUME_API_KEY> or an x-api-key header. Sume's docs prefer OAuth for interactive clients and keep API-key remote MCP for existing users and automation.

An API-key session sees Sume's full hosted tool set, write and paid tools included, and the key sits in mcp.json as plain text. Keep that file private, don't paste the key into a chat, and rotate it if it ever appears in logs or chat history.

Will LM Studio ask before a Sume tool runs?

Yes, according to LM Studio's 0.3.17 release notes. When a model calls a tool, LM Studio shows a confirmation dialog where you can review the arguments, and edit them, before the call runs. You can allow a tool once or always; an always-allowed tool skips the dialog from then on, and you can change that in App Settings > Tools & Integrations.

Sume's paid tools add their own gates. Each needs an idempotency_key, dry_run=true returns an admission and cost preview without submitting the job, and max_spend_usd caps a call only when it is sent. A workable split:

What are the limits?

• Sume's create tools return a job, not a finished file: the model waits with jobs_wait, which holds one call for at most 55 seconds, reads jobs_result, and never resubmits the paid create. MCP tool call timeouts on long-running video jobs covers the wait contract.
• LM Studio warns that some MCP servers designed for Claude, ChatGPT, or Gemini can use excessive tokens, bog down a local model, and trigger frequent context overflows. LM Studio's token count includes tool definitions, and an API-key session lists Sume's full hosted tool set, so watch your model's context.
• Hosted MCP cannot read files from your laptop, and reference images for generate_image must be public HTTPS URLs.
• Before a real prompt, ask the model to call mcp_health and tools_list; both are read-only. Sume MCP tools list groups every hosted tool by read, write, and paid.