ETSI Standards Connector

An MCP (Model Context Protocol) server that fetches, parses, and summarises ETSI telecommunications and media standards documents using Claude. Once registered, Claude can autonomously look up and ingest standards mid-conversation, then use the resulting summaries to answer questions, write conformant XML, or generate code that meets a specification.

How it works

You: "Does ETSI TS 102 818 define a required XML namespace?"

Claude --> search_standard("ETSI TS 102 818")
       --> ingest_standard("ETSI TS 102 818")   [downloads PDF, summarises with Claude Opus]
       --> reads standards/TS-102-818.md
       --> "Yes. Clause 5.2 [M] requires the namespace..."

The pipeline has three stages:

1. Search — queries the ETSI standards search API to find the current version of a standard and its PDF download URL.

2. Ingest — downloads the PDF, extracts the text, and calls Claude Opus to produce a structured summary. The summary captures all mandatory/optional requirements, data structures, workflows, allowed values, and normative references. It is written to standards/<id>.md and cached for future use.

3. Query — any subsequent question, code generation, or XML output task reads the local summary rather than the PDF, keeping token costs low.

Related MCP server: mcp-server-3gpp

Tools exposed over MCP

A resource template standards://{id} also exposes summaries directly as MCP resources.

Summary format

Every standards/<id>.md follows a fixed schema designed to be token-efficient and machine-readable:

• Scope — what the standard covers

• Key Concepts — terms with definitions

• Normative References — other standards required for conformance

• Data Structures — fields, types, allowed values; binary structures include byte offsets and hex values

• Rules and Constraints — numbered list, each prefixed [M] (mandatory) or [O] (optional); recommended values marked *

• Workflows and Sequences — numbered steps or Mermaid diagrams

• Conformance Requirements — all SHALL / SHOULD / MAY clauses

• Related Standards — standards that may warrant further ingestion

Requirements

• Node.js 20 or later

No Anthropic API key is required in the MCP server. Summarisation is performed by the Claude instance that is already running (Claude Code or Claude Desktop) — the server only handles fetching and parsing.

Installation

git clone <repo>
cd claude-etsi-connector
npm install
npm run build

Registering with Claude

Claude Code (global — available in all projects)

claude mcp add --scope user etsi -- node /absolute/path/to/claude-etsi-connector/dist/src/server.js

This writes to ~/.claude.json. No API key configuration is needed.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "etsi": {
      "command": "node",
      "args": ["/absolute/path/to/claude-etsi-connector/dist/src/server.js"]
    }
  }
}

Restart Claude Desktop after saving.

Verify

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1"}}}' \
  | node dist/server.js

You should see a JSON response containing "name":"etsi-connector".

Environment variables

Example usage

Once registered, Claude can use the tools directly. You can also prompt it explicitly:

"Ingest ETSI TS 103 720 and tell me what video codec profiles are mandatory."

"Using the summary for TS 102 818, write a valid XML service information document."

"What standards does TS 102 818 normatively reference that haven't been ingested yet?"

Project structure

claude-etsi-connector/
├── src/
│   ├── server.ts       # MCP server — tool and resource registration
│   ├── fetcher.ts      # ETSI search API + PDF download
│   ├── parser.ts       # PDF text extraction
│   ├── summariser.ts   # Claude API call → standards/<id>.md
│   └── config.ts       # Paths and model constants
├── prompts/
│   └── summarise.md    # Summarisation prompt template
├── standards/          # Generated summaries (committed to git)
├── cache/              # Downloaded PDFs (git-ignored)
└── dist/               # Compiled output (git-ignored)

Known limitations

• Summarisation calls claude-opus-4-6 and may take 30–90 seconds for a long specification.

• Very large documents (>150,000 characters of extracted text, roughly 300+ pages) are truncated. The summary will note where truncation occurred.

• PDF text extraction quality depends on how the PDF was produced. Scanned or image-based PDFs will yield poor results.

• The ETSI search API does not reliably flag superseded versions; the connector always takes the first (most recent) result.