Sozio Thin
Enables reproducible analysis of 100 curated Swiss open-data resources by searching profiles, materializing PXWeb data, validating and executing SQL, and formatting reproduction details, with reasoning delegated to the MCP client.
README
Sozio Thin
Sozio Thin is a standalone MCP server for reproducible analysis of 100 curated Swiss open-data resources. It runs locally without Vespa, Docker, Gemini, or an LLM API key. The MCP client (Codex or Claude Code) performs reasoning and SQL generation; the server only searches profiles, exposes context, materializes PXWeb data, validates SQL, executes DuckDB, and formats reproduction details.
Claude Desktop
For Claude Desktop on Windows, download
Sozio-Thin-...-Windows-x64.mcpb from the
latest GitHub release.
Then use one of these installation methods:
- Double-click the downloaded
.mcpbfile. - Drag the file into the Claude Desktop window.
- In Claude Desktop, open
Developer > Extensions > Install Extensionand select the file.
Claude Desktop shows an installation dialog and manages the local extension.
No JSON configuration, OAuth flow, remote server, or API key is required.
After installation, open + > Connectors in a chat and select Sozio Thin.
Windows installer for Codex and Claude Code
For Codex or Claude Code, use the Windows wizard from the latest GitHub release:
- Download
Sozio-Thin-Setup-...-Windows-x64.exe. - Run the installer without administrator rights.
- Select Codex and/or Claude Code integration.
- Restart the selected client and check for the
sozio-thinMCP server.
The installer includes Python, all runtime dependencies, and the 100-resource catalog. Git, uv, Python, Docker, and API keys are not required. The current installer is not code-signed, so Windows may show an "Unknown publisher" warning. Verify the SHA-256 checksum published in the release notes.
Requirements
The following requirements apply only to installation from source:
- Python 3.12 or newer
- uv
- Internet access when a selected public dataset must be read
The catalog contains profiles and source links, not copies of the datasets. Version 0.1 includes PXWeb, CSV, Parquet, and JSON resources. Spreadsheets are excluded because multi-row headers cannot be reproduced reliably by a generic DuckDB reader.
Setup
cd products/sozio-thin
uv sync
uv run sozio-thin doctor
uv run sozio-thin search "Bevölkerung nach Alter und Kanton"
Start the stdio MCP server:
uv run sozio-thin mcp-server
No environment variable is required. Runtime files are written below state/.
Set SOZIO_THIN_STATE_DIR to use another writable directory.
Slow PXWeb installations can be tuned with
SOZIO_THIN_PXWEB_TIMEOUT_SECONDS and SOZIO_THIN_PXWEB_RETRY_ATTEMPTS.
Codex configuration
Add this server to the Codex MCP configuration, replacing the path:
[mcp_servers.sozio-thin]
command = "uv"
args = [
"--directory",
"C:\\path\\to\\sozio-thin",
"run",
"sozio-thin",
"mcp-server",
]
The server uses the model already running in Codex. It cannot access or bill a Codex subscription itself.
Claude Code configuration
claude mcp add sozio-thin -- uv --directory C:\path\to\sozio-thin run sozio-thin mcp-server
Restart the client after changing MCP configuration.
Recommended workflow
- Call
search_resourcesfor one or more parts of the research question. - Call
get_context_bundlewith the chosen resource IDs. - Materialize every selected PXWeb resource with a narrow
scope. - Inspect live schemas and relevant values with
inspect_source. - Formulate joins only after checking grain and key coding.
- Call
validate_sql, thenexecute_sql. - Review the returned rows in the MCP client.
- Call
format_reproduction_bundle.
The server deliberately has no answer_question, LLM review, reranking, or
literature-research tool. Those steps belong to the MCP client.
Catalog
Every product-owned input is inspectable:
catalog/resources.json: local BM25 search documentscatalog/profiles/*.json: complete planning profilescatalog/topics.json: fixed topic assignments and quotascatalog/joins.json: suggested join pairscatalog/licenses.json: source attribution and license-review linkscatalog/selection.json: deterministic selection evidence
The source data remains governed by its publisher's terms. Verify the linked terms before redistribution or commercial reuse.
Catalog maintenance
The shipped catalog was generated from the full project's readiness report. To rebuild it when those private build inputs are available:
python scripts/select_resources.py `
--readiness-report C:\path\to\resource_readiness_report.json `
--profile-root C:\path\to\raw_profiles `
--output catalog
python scripts/verify_catalog.py
The runtime does not need those full-project inputs.
Tests
uv run pytest
uv run ruff check .
Online source probes are marked online and are not part of the default unit
test run.
Build the Windows installer with:
powershell -NoProfile -ExecutionPolicy Bypass -File scripts\build_windows_installer.ps1
The build also creates the official Claude Desktop .mcpb package and validates
its manifest with the pinned MCPB CLI.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.