The Atlas MCP Server
for AI that sees your energy book
We expose the ATLAS platform through a native Model Context Protocol (MCP) server. Plug it into Claude, GitHub Copilot, Cursor or any MCP-capable assistant and ask questions about your portfolio in plain language — answered against live platform data, not stale screenshots.
atlas_run_var(portfolio="TRADING", confidence=0.95)
What is the Model Context Protocol?
An open standard that lets AI assistants connect securely to external data and tools — instead of training on your data or pasting exports into a chat.
Through the protocol, an MCP server gives the assistant live, permissioned access to typed tools and resources.
Because ATLAS ships its own MCP server, every AI assistant your team already uses can become an ATLAS terminal — no bespoke connectors, no screen scraping, no data leaving your tenant beyond what you allow.
One server, every assistant
Claude, OpenAI, Copilot, Cursor and any MCP client speak the same protocol. Connect once, use everywhere.
Permissioned by design
Tools are permissioned and auditable. Read-only by default; write tools follow the authenticated user's role inside ATLAS.
Always live data
Every answer is computed on demand from the same Data Room your platform uses — never a stale copy.
From questions to actions
Query positions, run risk, preview scenarios and — with the right role — prepare bids, all from the chat.
What your team can ask
No prompts to learn, no dashboards to open — just ask and get answers grounded in live ATLAS data.
“Which of our trading books is most exposed to next month’s gas price move?”
“Summarise our PnL this week vs last week and explain the biggest driver.”
“Draft tomorrow’s Day-Ahead bids for the wind portfolio using the latest forecast.”
Secure by default, like the API
The Atlas MCP Server rides on the exact same enterprise controls as the ATLAS ETRM API — AI access never becomes a security back door.
Role-based access
Every tool call respects the authenticated user’s role in ATLAS. Read-only by default; write actions are opt-in per integration.
Short-lived tokens
Authenticated exactly like the API — a JWT from POST /api/token that expires after one hour and is refreshed automatically.
Full audit trail
Every MCP call is recorded with who, what, when and the data returned — the same audit log as the platform UI.
Tenant isolation
Strict multi-tenant isolation. An assistant only ever sees your tenant’s data, within the user’s role.
No data leaves your deployment beyond what the authenticated role allows — and every action stays in your audit log.
Works with the assistants you already use
Claude
Anthropic
OpenAI
GPT models
GitHub Copilot
VS Code & GitHub
Cursor
AI code editorAny MCP client
Open standardAtlas speaks the open MCP protocol, so it connects to any compliant client — today’s and tomorrow’s.
Put an ATLAS analyst in every chat
Give your team AI assistants that reason over real positions, risk and market data — with enterprise governance built in.