Zed MCP server: add Sume's hosted MCP in context_servers

To add an MCP server to Zed, open Settings → AI → MCP Servers, click Add Server, and choose Add Local Server for a command Zed runs, Add Remote Server for a URL, or Install from Extensions. Custom servers are saved under context_servers in your settings file. For Sume's hosted server, add a remote server whose url is https://mcp.sume.com/mcp: without an Authorization header, Zed signs you in through the standard MCP OAuth flow; with one, it sends your Sume API key instead.

Zed's side comes from its Model Context Protocol and Tool Permissions pages; Sume's from MCP quickstart, OAuth and API keys, MCP tools and gates, and Jobs and results, all read on 2026-09-27, plus current server code where noted. Sume has no Zed extension and no official connector: this is a plain remote MCP connection. Sume's basics page says hosted MCP still works but is not the primary integration path today.

How do I add Sume's hosted MCP server to Zed?

Open the MCP Servers page from Settings → AI → MCP Servers, or run the agent: open settings action and select MCP Servers. Click Add Server in the page header, then Add Remote Server. You can also edit the settings file directly (zed: open settings file). The minimal entry is a name and a URL:

{
  "context_servers": {
    "sume": {
      "url": "https://mcp.sume.com/mcp"
    }
  }
}

Should Zed sign in with OAuth or send an API key?

Sume accepts both. Leave the Authorization header out and Zed prompts you to authenticate through the standard MCP OAuth flow. On Sume's consent page, Read is locked on and Write is off by default. A read-only session sees only read tools: paid tools such as generate_image are hidden, and a call to one returns insufficient_scope. Turn Write on if Zed's agent should generate media. In current code a Sume OAuth token lasts one hour and Sume issues no refresh token, so expect to sign in again after that.

For an API key, add "headers": { "Authorization": "Bearer <SUME_API_KEY>" } to the entry. Sume also accepts an x-api-key header, but Zed's docs tie the OAuth prompt to a missing Authorization header, so use the Bearer form here. An API-key session sees the full hosted tool set, write and paid tools included. The key then lives in your Zed settings: never commit it, never paste it into chat, and rotate it if it appears in logs or chat history.

Once connected, the dot next to sume on the MCP Servers page turns green with the tooltip “Server is active”. The Zed Agent uses the server directly, and Zed can forward it to External Agents over the Agent Client Protocol.

How do I make Zed ask before a paid Sume tool runs?

In Zed v0.224.0 and later, agent.tool_permissions.default decides. confirm, the default, prompts before any tool action, MCP tool calls included; allow auto-approves; deny blocks every tool action. Per-tool rules name MCP tools as mcp:<server>:<tool_name>. For MCP tools, the per-tool default is the control to use, because pattern rules match against an empty string for them. A tool-specific default outranks the global one, so this keeps a paid tool prompting even if you later switch the global default to allow:

{
  "agent": {
    "tool_permissions": {
      "default": "confirm",
      "tools": {
        "mcp:sume:tools_list": { "default": "allow" },
        "mcp:sume:jobs_wait": { "default": "allow" },
        "mcp:sume:generate_video": { "default": "confirm" }
      }
    }
  }
}

Will Zed time out while a video renders?

Zed's MCP page documents no tool-call timeout for context servers, and Sume keeps each wait short: jobs_wait holds one call open for at most 55 seconds, 50 by default. When a wait ends with wait_slice_expired, the agent should call jobs_wait again with the same ids and never resubmit the paid create. MCP tool call timeouts on long-running video jobs has the full pattern.

How do I check that Sume is connected?

Ask the agent to call mcp_health, which confirms the endpoint, auth source, and safety posture, then tools_list for every tool the session can see. Zed notes that naming the server in your prompt can help the model pick its tools. When a tool call fails, Zed shows the server's error message in the agent's response; on an OAuth session, insufficient_scope there means Write is off, as Fix MCP insufficient_scope on Sume explains.