Definition
Plain language
Whether one tool's output format makes sense as another tool's input.
As stated in the literature
In Firefly's tool compatibility graph, the LLM-judged property of whether one API's response schema can plausibly feed into another's input schema.
Why it matters: Knowing which tools can chain into which lets training systems generate realistic multi-tool trajectories without testing every combination by hand.
For example, a tool that returns a list of flight IDs is token-compatible with a booking tool that takes a flight ID as input.