BlitFlow
Documentation
Run your AI workflows from the API, SDK, MCP, or CLI. One contract across every surface, reproducible every run.
blitflow run acme/generate-image \
--input prompt="a calm mountain lake at dawn"import { BlitClient } from "@blitflow/sdk";
const { run } = new BlitClient({ apiKey: process.env.BLITFLOW_TOKEN });
const outputs = await run("acme/generate-image", {
prompt: { value: "a calm mountain lake at dawn" },
});{
"mcpServers": {
"blitflow": { "url": "https://mcp.blitflow.com" }
}
}curl https://api.blitflow.dev/v1/runs \
-H "Authorization: Bearer $BLITFLOW_KEY" \
-d '{ "node": "acme/generate-image", "input": { "prompt": "a calm mountain lake at dawn" } }'Four ways to run the same workflow
Every published workflow speaks one contract. Pick the surface that fits. The inputs, outputs, and results are identical.
HTTP API
REST + SSE from any language. Start a run, stream outputs, poll status, no SDK required.
Read the reference →TypeScript SDK
A typed client, a workflow builder, and streaming helpers so runs feel like local function calls.
Read the reference →MCP server
Expose BlitFlow as tools for AI agents. Browse nodes, run workflows, read artifacts over MCP.
Read the reference →CLI
Drive runs from your terminal and wire BlitFlow into scripts and CI.
Read the reference →Ship your first run in minutes
Follow the quickstart to authenticate, start a run, and stream the output, then reach for the API, SDK, or MCP reference when you need the details.