Sofia Data MCP
MCP server for querying and analyzing urban data from the Sofia municipality CKAN portal. Provides tools for dataset search, retrieval, preview, summarization, and faceted browsing.
README
Sofia Data MCP
TypeScript MCP server, stdio bridge, and CLI client for the Sofia municipality urban data portal at https://urbandata.sofia.bg.
Architecture
- Canonical server transport: Streamable HTTP
- Desktop compatibility transport: local
stdiobridge that forwards to HTTP - Local default endpoint:
http://127.0.0.1:3000/mcp - Source portal integration: CKAN Action API
Connect an MCP Client
The deployed server is public, requires no auth, and speaks Streamable HTTP at:
https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/mcp
Clients with native remote/HTTP MCP support (recommended)
VS Code (Copilot Chat) — add to .vscode/mcp.json in your project (or
your user mcp.json via the MCP: Open User Configuration command):
{
"servers": {
"sofia-data": {
"type": "http",
"url": "https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/mcp"
}
}
}
Cursor — add to .cursor/mcp.json:
{
"mcpServers": {
"sofia-data": {
"url": "https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/mcp"
}
}
}
Clients that only support local stdio servers (e.g. Claude Desktop)
Claude Desktop's claude_desktop_config.json only launches local processes,
so bridge to the remote HTTP endpoint with the widely-used mcp-remote
adapter (no install/build required, runs via npx):
{
"mcpServers": {
"sofia-data": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/mcp"
]
}
}
}
Alternatively, this repo's own @sofia-data/stdio-bridge package works the
same way — point it at the deployed URL instead of running the server
locally:
pnpm --filter @sofia-data/stdio-bridge dev -- \
--upstream-url https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/mcp
Quick smoke test (no client needed)
curl -s https://sofia-data-mcp-oparjiwnqa-ew.a.run.app/health
Packages
@sofia-data/core: shared CKAN client, normalization, analytics, and MCP tool definitions@sofia-data/http-server: HTTP MCP server@sofia-data/stdio-bridge:stdiobridge to an HTTP MCP server@sofia-data/cli: bilingual CLI client
Requirements
- Node.js 24+
- Corepack available
Install
corepack enable
corepack prepare pnpm@10.8.1 --activate
pnpm install
Build
pnpm build
Run Locally
Start the HTTP server:
pnpm dev:server
Default endpoint:
http://127.0.0.1:3000/mcp
Run the CLI:
pnpm dev:cli
Run the stdio bridge:
pnpm dev:bridge
Override the bridge upstream URL:
pnpm --filter @sofia-data/stdio-bridge dev -- --upstream-url http://127.0.0.1:3000/mcp
Docker
Build and run with Docker Compose:
docker compose up --build
End-to-End Tests
packages/e2e runs a full black-box smoke test against a deployed
instance (health check, MCP initialize handshake, listTools, and a real
search_datasets call) — distinct from packages/http-server's unit tests,
which spin up a local in-process server. It's skipped automatically (not
failed) when no target is given, so it never affects normal pnpm test/CI:
E2E_BASE_URL=https://sofia-data-mcp-oparjiwnqa-ew.a.run.app pnpm test:e2e
This also runs automatically as a post-deploy gate in
.github/workflows/deploy.yml, against the URL of the revision that was
just deployed.
Deploying to Google Cloud Run
See deploy/gcp/README.md for scripts that deploy
http-server to Cloud Run (capped at 1 instance) with an automated budget
kill-switch that scales the service to zero if spend reaches a configured
threshold (default $2).
Environment Variables
Copy .env.example and adjust as needed.
Key values:
SOFIA_CKAN_BASE_URLHOSTPORTMCP_HTTP_PATHREQUEST_TIMEOUT_MS=30000PREVIEW_MAX_BYTES=262144MAX_SEARCH_RESULTS=50CACHE_TTL_MS=300000— in-memory TTL cache for CKAN reads (groups, organizations, search results, dataset lookups). Set to0to disable caching.ALLOWED_HOSTS/ALLOWED_ORIGINS— comma-separated extra hostnames/origins (beyond127.0.0.1/localhost) allowed to reach the MCP endpoint. Required when deploying publicly (e.g. Cloud Run) so the server's own DNS-rebinding/CORS protection doesn't reject requests to its own public hostname. Left unset, the server only accepts local traffic.MCP_UPSTREAM_URLMCP_SERVER_URL
Tools
search_datasetsget_datasetlist_groupsget_grouplist_organizationsget_organizationlist_dataset_resourcespreview_resourcesummarize_datasetfind_related_datasetsfacet_datasets— returns catalog-wide counts (group/organization/format/license) computed by CKAN's native faceting, not a client-side count over a capped page of results
Dataset Metadata Enrichment
Dataset summaries include a semanticProfile field generated generically from CKAN
metadata alone (groups, tags, resource formats) — there is no per-dataset hardcoded
knowledge to maintain, so it scales automatically as datasets are added or renamed on the
portal. For exact field-level structure, use summarize_dataset or preview_resource,
which parse an actual sampled resource (JSON/GeoJSON) to report real geometry types and
property/field names.
Group names that were corrupted by a source-side Cyrillic transliteration/encoding bug
(e.g. bnopa3hoo6pa3ne) are automatically repaired using a generic transliteration of the
group's title, rather than a hardcoded name-to-name mapping.
Testing
Each package has a unit test suite using Node's built-in test runner:
pnpm test
CI runs pnpm build, pnpm check (typecheck), and pnpm test on every push/PR to main
(see .github/workflows/ci.yml).
Automated dependency updates
Dependabot (.github/dependabot.yml) checks weekly for updates to npm/pnpm packages,
GitHub Actions, and the Docker base image, grouping patch/minor updates into a single PR
per ecosystem. .github/workflows/dependabot-auto-merge.yml automatically approves and
enables auto-merge for patch/minor version PRs; the merge only completes once the
required build-and-test CI check passes on that PR (enforced by branch protection on
main). Major version bumps are left for manual review.
Notes
- Technical naming is English-only.
- CLI display text is bilingual and rendered as two lines with English first.
- Source metadata from the portal is preserved as published.
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.