Decompose
Break long text into classified semantic units (authority, risk, attention) deterministically—no LLM—for safer agent preprocessing.
Overview
io.github.echology-io/decompose is a Ship-phase MCP server that deterministically decomposes text into authority, risk, and attention semantic units without an LLM.
What is this MCP server?
- Decomposes text into classified semantic units without calling an LLM
- Labels dimensions include authority, risk, and attention per description
- PyPI package decompose-mcp with stdio MCP transport
- Useful preprocessing step to cut token cost and add structure upstream of the model
- Registry version 0.1.1; GitHub source at echology-io/decompose
- PyPI identifier decompose-mcp version 0.1.1
- Transport: stdio
- Explicitly no LLM in server description
Community signal: 9 GitHub stars.
What problem does it solve?
You feed walls of text into the agent and get shallow summaries because nothing structured the document by risk and salience first.
Who is it for?
Builders who want cheap, repeatable text segmentation before LLM analysis of policies, specs, or support escalations.
Skip if: Teams that need full legal interpretation, embeddings search, or LLM-native summarization as the only step.
What do I get? / Deliverables
After adding the MCP server, the agent can preprocess text into labeled units for tighter review, scoping, and triage workflows.
- Classified semantic units with authority, risk, and attention facets
- Structured input for follow-on agent reasoning steps
- Lower-token workflows by segmenting before full LLM passes
Recommended MCP Servers
Journey fit
Ship is canonical because the highest-value moment is structured review of content, policies, and release artifacts before they go live. review matches pre-ship reading of docs, terms, and messaging where labeled segments beat monolithic prompts.
How it compares
Deterministic text-decomposition MCP, not a chat model or vector database integration.
Common Questions / FAQ
Who is io.github.echology-io/decompose for?
It is for developers who want MCP-accessible, non-LLM decomposition of documents for review, scoping, and structured agent prompts.
When should I use io.github.echology-io/decompose?
Use it before deep LLM analysis when you need authority, risk, and attention labels on long pasted text during review or validation.
How do I add io.github.echology-io/decompose to my agent?
Install decompose-mcp from PyPI, configure the stdio MCP entry in Claude Code, Cursor, or your host, and invoke decomposition tools on source text.