Supercharging Data Portals with the PortalJS MCP Server
Anuar Ustayev
Back in September this year, we published our first look at using MCP (Model Context Protocol) servers to give AI assistants structured access to data portals.
Now the implementation is live and fully open source.
PortalJS MCP runs in production on Cloudflare’s MCP SDK, which gives us a fast, global, edge-native runtime. It comes with low latency, high reliability, and no “AI integration infra tax” for you to pay.
The PortalJS MCP server is publicly available at:
mcp.portaljs.com
If your data portal runs on PortalJS Cloud, connecting it is dead simple. Your MCP endpoint is:
mcp.portaljs.com/@org-name/sse
Paste that into ChatGPT, Claude, or any MCP-capable client, and your AI assistant immediately gains structured access to your datasets, metadata, and previews.
And because we think this should be a standard building block for modern data portals, we’ve open sourced the whole implementation here:
https://github.com/datopian/portaljs-mcp-server
Use it, fork it, deploy your own version, or just read through it to understand how MCP can sit cleanly on top of a data portal.
Why MCP Is a Game-Changer for Data Portals
AI chats are powerful, but without structured access they’re basically guessing. MCP fixes that by giving models secure, predictable tools to interact with real systems — including your data portal.
In practice, this unlocks:
- Reliable dataset discovery backed by actual portal data search
- Accurate metadata exploration without hallucination risk
- On-demand previews (rows, schema, field types)
- One clean integration that works across multiple AI clients
This effectively turns your AI assistant into a precision data navigator — not just a polite autocomplete engine.
What’s Available in the MCP Today
The initial toolset focuses on high-value workflows for discovery and exploration:
Search tool enables data discovery
- List datasets
- Keyword search
- Metadata filtering
- Dataset summaries
Get tool for metadata exploration
- Resource lists
- Field definitions
- Schema inspection
- Full metadata extraction
Table preview
- First N rows
- Column summaries
- Type inference
- Lightweight profiling
These tools are designed to be fast, bounded, and safe. The model doesn’t pull full datasets — it gets structured previews that are ideal for reasoning and analysis.
Works with ChatGPT, Claude, VS Code, and More
Our MCP server is model-agnostic by default:
- Claude — native MCP support
- ChatGPT Desktop — native MCP support
- VS Code MCP clients — plug-and-play
- Future MCP-enabled tools — automatically compatible
Wherever your team uses AI, your portal can now show up as a first-class, tool-based data source.
Why Cloudflare’s MCP SDK?
We chose Cloudflare’s SDK because MCP should feel like infrastructure you never have to think about.
Using Cloudflare gives us:
- Edge deployment by default → fast globally, no region bottlenecks
- Battle-tested SSE support → stable streaming tool calls
- Simple scaling model → no infra babysitting as usage grows
This matters because AI tooling isn’t forgiving. If your MCP endpoint is slow or flaky, your user’s trust evaporates instantly. Cloudflare’s runtime lets us keep it sharp.
What’s Coming Next
This is only the first layer. We’re already expanding the MCP toolbox, including:
- Write-back tools (tags, notes, curation workflows)
- Automated metadata enrichment
- Data quality checks
- Permission-aware exploration
- Semantic search
- Lineage and observability integration
The direction is clear: your data portal becomes an intelligent interface, not a static catalog.
Try It Today
If your portal runs on PortalJS Cloud, your MCP endpoint is:
https://mcp.portaljs.com/@org-name/sse
Plug it into your AI assistant and start exploring your data conversationally — with real structure, real metadata, and real previews.
Want help rolling this out to your team or customers? Reach out. We’re building this to make data portals genuinely useful in an AI-first world.