A deterministic text analysis tool that decomposes documents into structured semantic units like risk, authority, and financial data without using an LLM.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install decompose-mcp
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install decompose-mcp using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Decompose MCP?
Decompose-mcp is a high-performance, deterministic utility designed to transform unstructured text into highly structured, actionable data. It serves as a critical cognitive primitive for any developer building with Openclaw Skills, allowing agents to understand the 'weight' of text through authority levels (e.g., mandatory vs. permissive) and risk categories (e.g., safety-critical vs. financial). Unlike LLM-based classifiers, this tool runs entirely locally, ensuring zero API costs, total privacy, and millisecond-level latency.
By integrating Decompose into your workflow, you can optimize your agent's context window by filtering out irrelevant content while retaining 100% of the semantic meaning required for decision-making. It is specifically tuned for technical specifications, legal contracts, and regulatory documents, identifying industry standards like ASTM, ISO, and OSHA with high precision.
Decompose MCP Use Cases
- Optimize context windows by 60-80% by pre-processing large documents before LLM ingestion.
- Automatically classify contract obligations and regulatory requirements based on RFC 2119 keywords.
- Extract specific industry standards, dates, and financial values from dense technical specifications.
- Prioritize content for human review or specialized agent chains using automated attention scoring.
- Build structured training data or compliance reports from raw markdown, HTML, or plain text files.
How Decompose MCP Works
- The tool receives input via the Model Context Protocol, accepting either raw text or a remote URL.
- It performs deterministic scanning to identify authority keywords such as "shall", "must", and "should".
- The engine calculates an attention score (0.0 to 10.0) by multiplying the authority weight by the identified risk multipliers.
- It extracts structured entities, including financial figures, milestones, and industry codes (ASTM, IBC, etc.).
- The output is delivered as a structured JSON object, allowing your AI agent to programmatically route or reason over the classified units.
Decompose MCP Setup
Install the package via pip or uv:
pip install decompose-mcp
Configure your environment to recognize the tool as part of your Openclaw Skills setup by adding the following to your MCP configuration file:
{
"mcpServers": {
"decompose": {
"command": "python3",
"args": ["-m", "decompose", "--serve"]
}
}
}
Verify the server is running correctly with a CLI test:
python3 -m decompose --text "The contractor shall provide all materials per ASTM C150-20."
Decompose MCP Data Schema & Taxonomy
The skill returns a JSON object containing a units array. Each semantic unit contains the following metadata taxonomy:
| Attribute | Description |
|---|---|
| authority | Classification of the obligation (mandatory, directive, permissive, etc.) |
| risk | The potential impact category (safety_critical, compliance, financial, etc.) |
| attention | A priority score from 0.0 to 10.0 based on authority and risk |
| actionable | A boolean indicating if the unit requires a specific human or agent action |
| entities | A list of referenced standards and codes (e.g., OSHA, ISO, ASCE) |
| financial | Extracted dollar amounts, percentages, and liquidated damages |
| heading_path | The hierarchical document path (e.g., Section 1.2 > Requirements) |
Decompose MCP Advanced Features
- Local-first processing ensures that sensitive data never leaves your machine during text analysis.
- SSRF protection logic validates DNS resolutions to block private and internal IP ranges for secure URL fetching.
- Deterministic output ensures that the same input always produces the exact same classification result.
- Compact mode support to reduce JSON payload sizes for highly efficient agent communication.
- High-throughput architecture capable of processing over 1,000 characters per millisecond on modern hardware.
SKILL.md
Loading
Related Openclaw Skills
JOULE DAO
A decentralized autonomous organization on the Base blockchain where agents earn JOULE tokens through verified productive work using Openclaw Skills.
Spark Bitcoin L2 Wallet Proxy
A secure HTTP API proxy that allows AI agents to interact with Spark Bitcoin L2 wallets using scoped bearer tokens without exposing private keys.
Spark Bitcoin L2 Proxy Deployer
A serverless middleware that wraps the Spark Bitcoin L2 SDK behind authenticated REST endpoints to give AI agents secure, scoped wallet access.
Team Communication
A specialized skill for seamless messaging and task coordination between AI agents within a structured team environment.
Little Steve Content Inbox
A lightweight content archiving and review tool for IM workflows that saves links, notes, and images directly from chat.
Boss: Automated BMAD R&D Pipeline Orchestrator
A high-level orchestrator that manages a team of specialized AI agents to handle the entire software development lifecycle from requirements to deployment.
METADATA
Featured*
