MCP tool description length: Claude Code 16,384 vs Sume
Claude Code 2.1.295 raised the tool-search description cut from 2,048 to 16,384 characters. What that means for a Sume MCP connection, and how to check it.

Claude Code 2.1.295 (Oct 8) now cuts MCP tool descriptions loaded through tool search at 16,384 characters. The old cut was 2,048. If you wrote a long description for your own MCP server, more of it now reaches the model; for Sume's hosted server at https://mcp.sume.com/mcp, the source defines a 2,048-byte limit, so the new cut should rarely matter.
What changed in Claude Code?
The change applies to descriptions that come in through tool search, not to every MCP tool. The changelog gives the two numbers and nothing else about wording or ranking, so do not assume the model reads a long description better than a short one.
| Item | Before | 2.1.295 |
|---|---|---|
| Description cut for tools loaded through tool search | 2,048 characters | 16,384 characters |
| Changelog date | Oct 8 2026 |
How long are Sume's tool descriptions?
Sume's MCP server package defines a 2,048-byte limit for tool descriptions and tests it. That is a source constant, not a promise about the live text. Read the live text with tools_schema for one tool name, or tools_list for all of them.
tools/list returns every tool in one response, so a client does not page through it. The server reports no listChanged push, so a client picks up new tools when it reconnects.
Does a higher cut change what you should do?
For Sume, rarely. The parts of a description that matter are the gates, and they are short: paid and write tools need idempotency_key, dry_run previews cost, and max_spend_usd is enforced only when you send it. Those stay the same at any cut.
If a client still truncates, the fix is the same as before: ask for the contract on demand.
- Call
tools_schemawith the tool name before the first paid call. - Keep OAuth on
mcp:readfor discovery and addmcp:writeonly for the session that submits. - Never rely on the description text for a limit. Send
max_spend_usdyourself.
What should you check after upgrading?
Run your normal discovery turn once on 2.1.295 and compare the tool list and one schema against the previous version. If the model now quotes a longer description, nothing on the Sume side needs to change. Details of the gates are in MCP tools and gates.
Sources
Related posts
More in Developers
- MCP tools/list repeating a cursor: Claude Code 2.1.295 and Sume
Claude Code 2.1.295 stops asking a server for the same tools page up to 20 times. Sume's tools/list returns one page with no nextCursor, so it never loops.
- Meta Marketing API: GenAI features vs an uploaded Sume file
Meta's Ad Creative Generative AI Terms cover GenAI features reached through the Marketing API. A script that only uploads a rendered file is a different step.
- Move an OpenRouter-style video client to Sume: the 400s to expect
Sume /v1/videos matches the OpenRouter video shape, but size, provider.options and seed return 400. The differences table and a Python asyncio polling client.
- MiniMax H3 disk space: 108 GB for SGLang, 19.5 GiB for INT8
How much disk does MiniMax H3 need? 108 GB for the SGLang route, 61.73 GiB BF16 safetensors, 19.53 GiB INT8. Why the numbers differ and what to budget.
Written by Sume