Claude MCP connector supports tool calls only; Sume MCP is tools-only

Anthropic's MCP connector supports only tool calls from the Model Context Protocol, per its documentation read on 2026-10-04. Prompts, resources and the rest of the spec are not used by the connector. That sounds like a limitation, but Sume's hosted MCP server at https://mcp.sume.com/mcp is itself tools-only, so the connector uses everything the server offers.

What does Sume declare?

In the server code on origin/main, the initialize result advertises a tools capability with listChanged set to false, under the server name sume-remote-mcp at version 0.4.0. The method handler answers four methods: initialize, ping, tools/list and tools/call. Anything else is a JSON-RPC method-not-found error (-32601). There are no prompts or resources to miss.

What does that mean for a Claude integration?

How does the server pair with the connector?

Both sides agree on tools, so the work is auth and sizing. The connector passes an authorization_token; you obtain it through Sume's OAuth flow or use an API key, as described on MCP OAuth and API keys. Remember that an API key gives the full tool set, while a read-only OAuth grant hides write and paid tools.

Because the server never pushes a changed tool list, re-list tools when you deploy a new client version rather than waiting for a notification. Use tools_list to see what the current session can call and tools_schema for one tool's contract, as the MCP tools and gates page advises.

What should I not assume?

Do not assume an MCP feature exists because the specification has it. The Sume docs say to use tools_list and tools_schema for the live contract and not to assume parity with the HTTP API. If your integration needs long-running work, use job tools: create, then jobs_wait, then jobs_result, as the Jobs and results page lays out.

If you want a connection check, mcp_health reports endpoint readiness, auth source and safety posture, and it works as a first call from the connector.

Does anything else change for Claude users?

Mostly the work is in naming and gating. Anthropic's connector lets you enable or disable tools per server through mcp_toolset, so you can expose three Sume tools without the other dozens. Keep that list short, because each tool description takes context, and Sume's descriptions are dense on purpose. Add the tools you need for one job at a time, then widen the list once the first workflow runs cleanly end to end.