branch_thought
Create a new branch of thinking from an existing thought. Returns only metadata, not full content.
LABSCO SUMMARY
The ideas live on disk; only their labels enter the conversation.
The interesting design choice is not the prompting, it is the storage. Search returns ids and metadata, the full text sits behind a resource URI, and filtering happens on the server — so a session with a hundred branches does not force a hundred paragraphs into context on every query. Directions are enum-typed (more_extreme, opposite, tangential), which makes branching something you steer rather than describe.
What it is
An MCP server for ideation that deliberately pushes against the obvious answer. Each thought is stored on disk as a resource at thought://[thoughtId] and tools return metadata plus a link, so the model pulls the full text of a thought only when it decides it needs it.
What you get
- A new unconventional thought on a problem, which can build on an earlier thought or rebel against it, returned as a
resource_linkplus structured fields rather than a text blob —generate_unreasonable_thought - A branch off an existing thought in a named direction —
more_extreme,oppositeortangential— as its own linked resource —branch_thought - Metadata search filtered by branch, by whether a thought was a rebellion, and by whether it challenges an assumption, with a limit on how many come back —
search_thoughts - Every thought readable on demand at
thought://[thoughtId], with rebellion status, challenged assumption, timestamp and branch information in the metadata - Thoughts written to a
.thoughts/directory, so a session's tree survives restarts and can be inspected as plain files
Requirements
No account and no key. The npm package name is unreasnable-thinker-server (0.3.0) and it is marked private, so build from a clone with npm install and npm run build, then point your client at the built index.js. It targets the MCP 2025-11-25 spec through @modelcontextprotocol/sdk v1.27.1, using tool title, annotations, outputSchema, structuredContent and resource_link — a client that ignores resource links will still work, but will have to fetch thought text itself.
Setup effort
One command — npx -y github:stagsz/Unconventional-thinking
USE CASES
01
Getting past the first obvious answer
Ask for an unreasonable thought on the problem, then branch it in the opposite direction to see what the inverse assumption implies.
02
Stress-testing a plan you already like
Generate thoughts that rebel against the current approach, then filter for the ones flagged as challenging an assumption and read only those.
03
Exploring one idea to its limit
Branch repeatedly in the more_extreme direction and follow the chain, so the escalation is recorded rather than reconstructed from the transcript.
04
Coming back to a brainstorm days later
Search the stored thoughts by branch to rebuild where you were, and open the full text of only the ones still worth reading.
3 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
generate_unreasonable_thought
Generate a new unreasonable thought that challenges conventional thinking. Efficiently creates thoughts without loading full context.
search_thoughts
Search for thought IDs by metadata. Returns only matching IDs and metadata, not full content. Enables efficient filtering.
BEFORE YOU INSTALL
What to know first
Tools you get
3 tools when we connected.
Requires
npx on PATH
CONTEXT COST
Light Standing cost
Rated by the size of the 3 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Add the block to your client, or run the command directly.
Server setup
Configuration file
JSON
{
"mcpServers": {
"Unconventional-thinking": {
"command": "npx",
"args": [
"-y",
"github:stagsz/Unconventional-thinking"
]
}
}
}npx
TERMINAL
npx -y github:stagsz/Unconventional-thinking
