Gemini CLI MCP servers not loading in an untrusted workspace
Gemini CLI v0.59.0 filters mcpServers in restricted mode. If Sume is missing in an untrusted folder, trust is the cause; then tell it from an OAuth issue.

If Sume's MCP server does not load in Gemini CLI, check whether the folder is trusted before you touch the Sume config. Gemini CLI v0.59.0 (2026-09-08) fails closed on workspace trust and filters mcpServers in restricted mode, so a server defined for a folder you have not trusted is simply not loaded, however correct the URL is.
That is Gemini CLI behavior, taken from its changelog read 2026-09-29; the changelog gives no more detail, so follow Gemini CLI's own documentation for how to trust a folder. The Sume side is short and is below.
How do I tell trust from a Sume problem?
Two different failures look alike from the chat window: the server never appears, or it appears and then a call is refused. Only the second one involves Sume.
| Symptom | Owner | Next step |
|---|---|---|
| Sume server never shows up in an untrusted folder | Gemini CLI workspace trust (v0.59.0 filters mcpServers in restricted mode) | Trust the folder, then start a new session |
| Server loads but sign-in never completes | OAuth login for the configured server name | Run the client's OAuth login for that server |
| Server loads, read calls work, a write is refused | Sume OAuth scope: mcp:read only | Grant Write at consent, or use an API-key session |
What should the Sume server entry point at?
Sume's quickstart says to point a streamable HTTP MCP server at https://mcp.sume.com/mcp and to run that client's OAuth login for the configured server name. The client should discover Sume protected-resource metadata from the MCP endpoint and send you to https://mcp.sume.com/oauth/authorize, which continues to https://mcp.sume.com/oauth/consent.
How do I confirm it loaded after trusting the folder?
Ask the agent to call a read-only tool. The quickstart's verify step uses tools_list, which lists every tool visible to the session, and mcp_health, which confirms the endpoint, auth source and safety posture. If tools_list answers, trust and sign-in are both fine.
Call the Sume MCP tool tools_list and summarize the available tools. Live ids
use underscores (tools_list, generate_image). Dotted aliases also work.Why can a write call fail after everything loads?
Default hosted OAuth grants read-only access (mcp:read). Write tools such as jobs_cancel or assets_create, and paid tools such as generate_image, return insufficient_scope until you grant mcp:write on consent, use an API-key remote MCP session, or use the Developer API. See OAuth and API keys.
Does Sume need any change for a trusted folder?
No. The Sume side of the setup does not depend on the folder: the same https://mcp.sume.com/mcp entry and the same OAuth login apply wherever the client runs. Once trust lets the entry load, follow the documented read-only playbook: connect with OAuth, leave Write off unless you need mutations, call mcp_health, and confirm authenticated.auth_source is mcp_oauth.
Then call tools_list and keep only the read_only tools in mind while Write is off. If Write was off, stop before mutating tools, because they return insufficient_scope.
Sources
Related posts
More in Developers
- Golang HTTP client retry: what net/http retries on POST
Go's http.Transport retries only network errors on reused connections, and a POST only with an Idempotency-Key header. Write the 429 and 5xx loop yourself.
- GPT-6 Astra tool calling needs the Responses API: meaning for Sume
OpenAI says GPT-6 Astra supports Chat Completions but its tool calling requires Responses. Use the Responses mcp tool for Sume, not a chat.completions loop.
- How long does GPT Image 2.5 take? Time your own jobs on Sume
Sume publishes no per-model image time for GPT Image 2.5, only a 30-second wait budget. A short curl script times your own prompts by quality and size.
- GPT Image 2.5 arena: run your own blind test through the API
Arena leaderboards rank images on other people's prompts. Here is a blind three-way test of Flare, Sunburst and GPT Image 2 on yours, with one script on Sume.
Written by Sume