One connector.
Any agent that needs to act on a market.
Sybilion's MCP connector gives any agent access to the same causal model that powers the API - forecasts, their drivers, and alerts - without plumbing, without a custom integration, and without changing the workflow the agent already runs.
Not a data feed. The reasoning layer behind the decision.
Most integrations surface a number. The Sybilion MCP connector surfaces what is moving the number, with what lag, and whether to act yet. That is the difference between an agent that reports and an agent that decides.
drivers
What moves what, and when
Every forecast returns its ranked external drivers - the macroeconomic indicators, other series the model found most associated with your own data - with feature-importance scores and Granger lag relationships that show the connection before the move.
get_forecast_artifact
forecasts
A forecast the agent can defend
Monthly series in, monthly forecast out, with quantile bands, per-driver attribution, and backtest metrics returned as retrievable artifacts. Every output shows its reasoning - so the agent's recommendation survives the room where it gets signed off.
submit_forecast → get_forecast → get_forecast_chart
alerts
What is moving against you
Pass your series context. Get back the related series that have moved significantly right now, ranked by relevance, each with its percentage change and the news behind it - synchronously, in the same session, with no additional configuration.
get_alerts
Forecasts and alerts accept optional region and category filters from the catalog - browse the valid ids with list_regions and list_categories. Every claim here maps to a tool you can call today. See the full schema in the docs.
One block. Any agent that speaks MCP.
No custom plumbing. No SDK to install. Add the connector, point your agent at it, and ask in plain language.
mcp
json{
"mcpServers": {
"sybilion": {
"url": "https://mcp.sybilion.dev/mcp"
}
}
}
# Authentication is OAuth: the client opens a browser login the first time you connect and you approve access - there is no API key to paste into the config.
# Then ask, in plain language:
"What drives EU polyethylene, and should we commit Q3 volume now?"
"Is now the right time to lock Q3 freight volume, or do the signals say wait?"
"What macroeconomic factors are turning relevant to our carbon allowance exposure this quarter?"
Forecast jobs are asynchronous - submit returns a job id, the agent polls get_forecast until the job completes, then retrieves the chart and artifacts. The connector paces the polling for you so the agent does not have to guess how long to wait.
Built to sit inside the agent you already have.
The Sybilion MCP connector does not ask your agent to change what it does. It gives it the one thing a foundation model cannot reach - a causal, lag-aware understanding of what is moving the market in front of it.
ChatGPT
Market decisions inside your ChatGPT workspace
Feed validated drivers and forecast bands straight into your positions and risk models, through the API or webhooks. No product to adopt. No change to the workflow.
Setup: Add mcp.sybilion.dev/mcp to your ChatGPT MCP server configuration. No code required.
Claude
A procurement agent that reasons about markets
Which external signals predict the series in front of it, and how confident to be. Your agent executes the task. Sybilion tells it which call to make first. One MCP connector, no plumbing.
Setup: Add mcp.sybilion.dev/mcp to your Claude MCP server configuration. Works in Claude.ai, Claude Desktop and even Claude Code.
TradingView
Causal context on top of the chart you already trust
Enrich the data you already sell or surface with the causal context (what drives the number, and when) that moves it from a chart to a decision your users can act on.
Setup: Add mcp.sybilion.dev/mcp to your TradingView Remix MCP configuration. Works alongside your existing indicators and data feeds.
A small, honest surface area.
The same model, reachable however your agent already works.
Transparent about what is in the connector today: Sybilion supports monthly business time series. Sub-monthly frequencies are on the roadmap, not in the public connector yet. We would rather tell you that than have you find out mid-agent-run.
What a connected agent actually looks like.
These are real tool call sequences. The agent asks in plain language. The connector handles the rest.
start building