# @modelcontextprotocol/server-sequential-thinking ## 2.1.1 ### Patch Changes - [#5435](https://github.com/modelcontextprotocol/servers/pull/5435) [`8da06bc`](https://github.com/modelcontextprotocol/servers/commit/8da06bc446aff78ff8829534aa3dc78d12b44c8b) [@cliffhall](https://github.com/cliffhall) - Bound the thought box logged to stderr. Since the multi-line box, every row was padded to the thought's longest line, so a thought with one long line and many short ones logged their product: a 60 KB thought wrote 400 million characters and could exhaust the server's memory. The box is now at most 100 characters wide (longer lines, and a long branch ID in the header, wrap) and draws at most 1000 rows of thought text (it then says how many lines it left out), so the log no longer grows with the thought's line count times its longest line. The thought kept in the server's history and the tool result are unchanged. ## 2.1.0 ### Minor Changes - [#5397](https://github.com/modelcontextprotocol/servers/pull/5397) [`93f89b1`](https://github.com/modelcontextprotocol/servers/commit/93f89b1ecb1443714b28cdb542969363bb932fcb) [@cliffhall](https://github.com/cliffhall) - Bound the thought history each connection keeps: at most `MAX_THOUGHT_HISTORY` thoughts (default 1000) are retained, the oldest dropped first and from its branch too, so a long-lived connection no longer grows memory without bound. ### Patch Changes - [#5361](https://github.com/modelcontextprotocol/servers/pull/5361) [`9a1cdf3`](https://github.com/modelcontextprotocol/servers/commit/9a1cdf38348b3ee0752e48ea61295889a11b63b0) [@cliffhall](https://github.com/cliffhall) - Draw a multi-line thought one row per line in the stderr thought box, sized to the longest line, instead of as one broken row ([#5205](https://github.com/modelcontextprotocol/servers/issues/5205)) - [#5352](https://github.com/modelcontextprotocol/servers/pull/5352) [`59b7f9d`](https://github.com/modelcontextprotocol/servers/commit/59b7f9d15150217217997f6b523cf8afeca1e0f9) [@cliffhall](https://github.com/cliffhall) - Remove the unused `yargs` dependency (and `@types/yargs`), so installing the server no longer pulls in yargs and its transitive packages. - [#5374](https://github.com/modelcontextprotocol/servers/pull/5374) [`c4d0cc4`](https://github.com/modelcontextprotocol/servers/commit/c4d0cc4e0041f72f9f05c6eb16fe818753f9aeac) [@cliffhall](https://github.com/cliffhall) - Trim the `sequentialthinking` tool description from 2,781 to 1,009 characters (about 566 to 198 tokens), so it fits the 1,024-character cap OpenAI and Azure OpenAI put on tool descriptions ([#799](https://github.com/modelcontextprotocol/servers/issues/799)). The "Key features" and "Parameters explained" sections are gone: each parameter is still described by the input schema, which is unchanged, and the guidance those field descriptions do not carry is kept ([#4507](https://github.com/modelcontextprotocol/servers/issues/4507)). ## 2.0.0 ### Major Changes - [#5118](https://github.com/modelcontextprotocol/servers/pull/5118) [`5b400a6`](https://github.com/modelcontextprotocol/servers/commit/5b400a6ec05bd883522665ebeae110bf1a2744ff) [@cliffhall](https://github.com/cliffhall) - Move to the TypeScript SDK v2 packages (`@modelcontextprotocol/server`) from `@modelcontextprotocol/sdk` 1.x ([#4856](https://github.com/modelcontextprotocol/servers/issues/4856)). Breaking for clients: the server now requires Node.js 24 or later; over stdio, a single incoming JSON-RPC message larger than 10 MiB now closes the connection (the SDK v2 default read-buffer cap; SDK 1.x buffered without a limit); an unknown tool name is rejected with JSON-RPC error `-32602` instead of returning an `isError` result; error messages no longer carry the `MCP error : ` prefix, and input-validation messages name the field first (`field: message`); tool schemas in `tools/list` declare JSON Schema 2020-12 instead of draft-07; and tools no longer advertise `execution.taskSupport`. ### Minor Changes - [#5131](https://github.com/modelcontextprotocol/servers/pull/5131) [`419200b`](https://github.com/modelcontextprotocol/servers/commit/419200b8a4d1e66c294575930ac7be29d4b37330) [@cliffhall](https://github.com/cliffhall) - Serve the 2026-07-28 protocol revision alongside the 2025 era on stdio, through the SDK's `serveStdio` entry ([#4852](https://github.com/modelcontextprotocol/servers/issues/4852)). The thought history lasts as long as the client's connection, on either era. ### Patch Changes - [#5144](https://github.com/modelcontextprotocol/servers/pull/5144) [`d6fd666`](https://github.com/modelcontextprotocol/servers/commit/d6fd666467c4615828067f3bc2ab4e6d08c3e4a3) [@cliffhall](https://github.com/cliffhall) - Pad the header row of the thought box logged to stderr to the border's width, so its right edge lines up with the rest of the box ([#5108](https://github.com/modelcontextprotocol/servers/issues/5108)). ## 1.0.0 The first semantic version, resuming semver after the date-stamped releases (last `2026.8.31`). It describes everything since `2026.8.31`, and is the last release of the legacy-era server, on the 1.x MCP SDK. ### Patch Changes - [#5015](https://github.com/modelcontextprotocol/servers/pull/5015) [`8a4139c`](https://github.com/modelcontextprotocol/servers/commit/8a4139c3d71b66e95d2f7cd2bc485d4781ecf1d7) [@cliffhall](https://github.com/cliffhall) - Advertise the `sequentialthinking` tool as `readOnlyHint: false` and `idempotentHint: false`. Every call appends to the server's thought history, and a call with both `branchFromThought` and `branchId` also appends to that branch, so the tool was never read-only or idempotent. `destructiveHint` and `openWorldHint` stay `false`. ([#4721](https://github.com/modelcontextprotocol/servers/issues/4721)) - [#4970](https://github.com/modelcontextprotocol/servers/pull/4970) [`86806cc`](https://github.com/modelcontextprotocol/servers/commit/86806cc3870a112f56870e86cd36bae8239edb02) [@cliffhall](https://github.com/cliffhall) - Build the server with an exported `createServer()` factory and attach stdio only when `index.js` runs as the binary, so the server can be tested in-process. No change to the tool, its schemas or its results. - [#5031](https://github.com/modelcontextprotocol/servers/pull/5031) [`bc52d0e`](https://github.com/modelcontextprotocol/servers/commit/bc52d0e07a1d9dca505746165b30f4c99f8cc79b) [@cliffhall](https://github.com/cliffhall) - The thought log on stderr no longer prints "undefined" in its headers: a revision with no `revisesThought` reads "Revision N/M", and a branch with no `branchId` reads "(from thought N)" with no ID. The box border is sized from the header's visible text, so colour escape codes no longer make it wider than its text. - [#5053](https://github.com/modelcontextprotocol/servers/pull/5053) [`c63255a`](https://github.com/modelcontextprotocol/servers/commit/c63255ae9616b110d6089fa343fcfdf3f5def534) [@cliffhall](https://github.com/cliffhall) - Stop shipping the compiled test helper `dist/__tests__/helpers.js` in the published package: the build now excludes everything under `__tests__/`, not just `*.test.ts`. - [#5036](https://github.com/modelcontextprotocol/servers/pull/5036) [`b5c5ccc`](https://github.com/modelcontextprotocol/servers/commit/b5c5cccb79a5d43b41247065fb9efd174363e799) [@cliffhall](https://github.com/cliffhall) - A `branchId` that names an `Object.prototype` key, such as `"constructor"` or `"__proto__"`, now creates and lists its branch like any other id instead of failing the call. A call that fails no longer adds its thought to the history first, so `thoughtHistoryLength` counts only thoughts that were accepted. - [#5087](https://github.com/modelcontextprotocol/servers/pull/5087) [`09edf63`](https://github.com/modelcontextprotocol/servers/commit/09edf633b93cdaf094ba43622294d587afcd1023) [@cliffhall](https://github.com/cliffhall) - Require `@modelcontextprotocol/sdk` 1.31.0 or later (was `^1.30.0`), so an install can no longer resolve an SDK version affected by GHSA-6qxp-vccf-f47h. The advisory concerns the SDK's OAuth client, which this server does not use.