Zero API credits
The run is driven by your seat's entitlement, not a pay‑per‑token key you have to top up.
You already pay for Claude Pro or Max. DISPATCH lets that subscription drive the canvas directly — over MCP and a sub‑authed CLI — instead of metering every node against pay‑as‑you‑go API credits. No API key. No second bill.
DISPATCH publishes the graph over MCP. Your subscription‑authed CLI attaches as the client, executes each node under your existing Claude Pro/Max entitlement, and writes results back. The canvas never sees a key, because there isn't one.
Since it's just another transport, funding is per node. An Investigate node can run on your Max seat while Create hits a local model and Evaluate stays on a metered API — same graph, three sources, one Run.
Zero API credits
The run is driven by your seat's entitlement, not a pay‑per‑token key you have to top up.
Canvas over MCP
Nodes, prompts, tools and context are published as an MCP surface the CLI can drive.
Sub-authed CLI
You sign in once with the CLI you already use; DISPATCH hands it nodes to execute.
Per-node funding
Subscription, local, or metered API — choose per node. One graph, mixed sources.
Switch the canvas between a metered API key and your Claude Max subscription, then run the chain. Watch the meter: same nodes, different bill.
The subscription runtime drives the run end‑to‑end. No metered credits are drawn.
No key vault, no proxy reselling tokens. Three steps, once per session.
DISPATCH publishes the graph as an MCP server: each node's prompt, allowed tools (web_search, read_file, write_file) and the shared Context bus.
Your Claude CLI signs in once with its own subscription session and attaches as the MCP client. No API key is generated, stored, or passed to the canvas.
On Run, the canvas hands each node to the CLI, which executes it under your entitlement and writes the output back downstream — exactly like a metered node would.
Set provider + model per node and swap live, mid-run.
openRoblox, Linear, Notion, Vercel, Safentic, Netlify — and more.
openOllama, llama.cpp, LM Studio — fully local, alongside the cloud.
open