Decompose MCP

1---
2name: decompose-mcp
3description: Decompose any text into classified semantic units — authority, risk, attention, entities. No LLM. Deterministic.
4metadata: {"clawdbot":{"emoji":"🧩","requires":{"anyBins":["python3","python"]},"install":[{"id":"pip","kind":"uv","pkg":"decompose-mcp","bins":["decompose"],"label":"Install decompose-mcp (pip/uv)"}]}}
5---
6 
7# Decompose
8 
9Decompose any text or URL into classified semantic units. Each unit gets authority level, risk category, attention score, entity extraction, and irreducibility flags. No LLM required. Deterministic. Runs locally.
10 
11## Setup
12 
13### 1. Install
14 
15```bash
16pip install decompose-mcp
17```
18 
19### 2. Configure MCP Server
20 
21Add to your OpenClaw MCP config:
22 
23```json
24{
25  "mcpServers": {
26    "decompose": {
27      "command": "python3",
28      "args": ["-m", "decompose", "--serve"]
29    }
30  }
31}
32```
33 
34### 3. Verify
35 
36```bash
37python3 -m decompose --text "The contractor shall provide all materials per ASTM C150-20."
38```
39 
40## Available Tools
41 
42### `decompose_text`
43 
44Decompose any text into classified semantic units.
45 
46**Parameters:**
47- `text` (required) — The text to decompose
48- `compact` (optional, default: false) — Omit zero-value fields for smaller output
49- `chunk_size` (optional, default: 2000) — Max characters per unit
50 
51**Example prompt:** "Decompose this spec and tell me which sections are mandatory"
52 
53**Returns:** JSON with `units` array. Each unit contains:
54- `authority` — mandatory, prohibitive, directive, permissive, conditional, informational
55- `risk` — safety_critical, security, compliance, financial, contractual, advisory, informational
56- `attention` — 0.0 to 10.0 priority score
57- `actionable` — whether someone needs to act on this
58- `irreducible` — whether content must be preserved verbatim
59- `entities` — referenced standards and codes (ASTM, ASCE, IBC, OSHA, etc.)
60- `dates` — extracted date references
61- `financial` — extracted dollar amounts and percentages
62- `heading_path` — document structure hierarchy
63 
64### `decompose_url`
65 
66Fetch a URL and decompose its content. Handles HTML, Markdown, and plain text.
67 
68**Parameters:**
69- `url` (required) — URL to fetch and decompose
70- `compact` (optional, default: false) — Omit zero-value fields
71 
72**Example prompt:** "Decompose https://spec.example.com/transport and show me the security requirements"
73 
74## What It Detects
75 
76- **Authority levels** — RFC 2119 keywords: "shall" = mandatory, "should" = directive, "may" = permissive
77- **Risk categories** — safety-critical, security, compliance, financial, contractual
78- **Attention scoring** — authority weight x risk multiplier, 0-10 scale
79- **Standards references** — ASTM, ASCE, IBC, OSHA, ACI, AISC, AWS, ISO, EN
80- **Financial values** — dollar amounts, percentages, retainage, liquidated damages
81- **Dates** — deadlines, milestones, notice periods
82- **Irreducibility** — legal mandates, threshold values, formulas that cannot be paraphrased
83 
84## Use Cases
85 
86- Pre-process documents before sending to your LLM — save 60-80% of context window
87- Classify specs, contracts, policies, regulations by obligation level
88- Extract standards references and compliance requirements
89- Route high-attention content to specialized analysis chains
90- Build structured training data from raw documents
91 
92## Library: `filter_for_llm()`
93 
94When using Decompose as a Python library (not MCP), `filter_for_llm()` pre-filters decompose output for LLM context windows:
95 
96```python
97from decompose import decompose_text, filter_for_llm
98 
99result = decompose_text(document_text)
100filtered = filter_for_llm(result, max_tokens=4000)
101# filtered["text"] = high-value units only (mandatory, safety-critical, financial, compliance)
102# filtered["meta"]["reduction_pct"] = typically 60-80% token reduction
103```
104 
105Keeps mandatory, prohibitive, directive, conditional authority + safety-critical, compliance, financial, contractual risk + requirement, constraint, data, definition types. Configurable via keyword arguments.
106 
107## Performance
108 
109- ~14ms average per document on Apple Silicon
110- 1,000+ chars/ms throughput
111- Zero API calls, zero cost, works offline
112- Deterministic — same input always produces same output
113 
114## Security & Trust
115 
116**Text classification is fully local.** The `decompose_text` tool performs all processing in-process with no network I/O. No data leaves your machine.
117 
118**URL fetching performs outbound HTTP requests.** The `decompose_url` tool fetches the target URL, which necessarily involves network I/O to the specified host. This is why the skill declares the `network` permission in `claw.json`. If you do not need URL fetching, you can use `decompose_text` exclusively with no network access required.
119 
120**SSRF protection.** URL fetching blocks private/internal IP ranges before connecting: `0.0.0.0/8`, `10.0.0.0/8`, `100.64.0.0/10`, `127.0.0.0/8`, `169.254.0.0/16`, `172.16.0.0/12`, `192.168.0.0/16`, `::1/128`, `fc00::/7`, `fe80::/10`. The implementation resolves the hostname via DNS *before* connecting and checks all returned addresses against the blocklist. See [`src/decompose/mcp_server.py` lines 19-49](https://github.com/echology-io/decompose/blob/main/src/decompose/mcp_server.py#L19-L49).
121 
122**No API keys or credentials required.** No external services are contacted except when using `decompose_url` to fetch user-specified URLs.
123 
124**Source code is fully auditable.** The complete source is published at [github.com/echology-io/decompose](https://github.com/echology-io/decompose). The PyPI package is built from this repo via GitHub Actions ([`publish.yml`](https://github.com/echology-io/decompose/blob/main/.github/workflows/publish.yml)) using PyPI Trusted Publishers (OIDC), so the published artifact is traceable to a specific commit.
125 
126## Resources
127 
128- [Source Code (GitHub)](https://github.com/echology-io/decompose) — full source, auditable
129- [PyPI](https://pypi.org/project/decompose-mcp/) — published via Trusted Publishers
130- [Documentation](https://echology.io/decompose)
131- [Blog: When Regex Beats an LLM](https://echology.io/blog/regex-beats-llm)
132- [Blog: Why Your Agent Needs a Cognitive Primitive](https://echology.io/blog/cognitive-primitive)
133