Give AI agents structured access to forecasting
The TSFM.ai API provides tool-calling-compatible endpoints that let agents select models, execute forecasts, and parse results without custom glue code.
{
"name": "forecast",
"arguments": {
"model": "amazon/chronos-bolt-base",
"series": [428, 435, 441, 438, 446, 452],
"horizon": 14,
"quantiles": [0.1, 0.5, 0.9]
}
}Agents call the same API surface through tool definitions or MCP.
Discovery
Agents can list and filter models programmatically
Execution
Structured JSON in, structured JSON out — no parsing hacks
Integration
Works with tool calling, MCP, and direct HTTP
Why agents need a forecasting API
LLMs can reason about time series problems but cannot run inference on foundation models. A structured API bridges that gap.
Structured model discovery
Agents can list available models, filter by capability, and select the best fit for a given task without hardcoded model IDs.
Predictable request and response shapes
The API returns typed JSON that agents can parse directly. No HTML scraping, no unstructured text, no ambiguous outputs.
MCP and tool-calling support
Use the API through MCP server definitions or OpenAI-compatible tool schemas. Both paths use the same underlying endpoints.
View MCP serverHow to connect an agent to the forecast API
Most agent frameworks support tool definitions or MCP servers. Pick the path that fits your stack.
- 1
Register a tool or MCP server
Define the forecast endpoint as a tool in your agent framework, or point your MCP client at the TSFM.ai MCP server.
- 2
Let the agent discover models
The /v1/models endpoint returns available models with metadata. Agents can filter by context length, latency, or task type.
- 3
Execute forecasts in the agent loop
The agent calls the forecast tool with series data and a model ID. Results come back as structured JSON with point forecasts and quantiles.
Frequently Asked Questions
Connect your agent to real forecasting
Give your agent access to production-grade time series models through a single API key.