TCT
Tool-Call Trace Format · A proposed standard format for serializing the trace of an agent's tool-call activity for audit and reproducibility.
Status
DraftVersion
Updated
Steward
Specification: https://agentic.review/standards/tool-call-trace/
Editorial note
Our own proposal. We do not generally propose standards — we cover them — but this gap is large enough that we wrote a draft to put a position on the table.
Our analysis
TCT is the only entry on this tracker that the publication itself drafted. We do not as a rule propose standards — we cover them — but the gap in the audit-trail layer is large enough, and the consequences of every vendor shipping a different incompatible trace format severe enough, that we put a position paper into the working-group conversation. The proposal is narrow: a JSON-Lines format for serializing the sequence of tool calls an agent made on a task, with explicit fields for the principal, the agent ID (deferring to Agent ID for the identifier scheme), the tool invoked, the arguments, the result, and a structured outcome (success / handled-failure / unhandled). The format is meant to be a write-target for runtimes and a read-source for audit tooling; it is not meant to be a query language or a storage format. We publish it here because we want it critiqued. Editor's note: this entry is the most likely on the tracker to be deprecated within a year — either because the working groups absorb a version of the format into MCP or Agent ID, or because vendor-specific formats settle into a de facto standard before a portable format takes root.
What we are watching
- Whether MCP or Agent ID absorbs a trace primitive
- Vendor-specific trace formats shipping in 2026 H2
- Working-group response to the draft proposal
How to cite this entry
Citation format: The Agentic Review. "TCT — Tool-Call Trace Format." Standards Tracker, last
updated 2026-02-26. https://agentic.review/standards/tool-call-trace/.
See our citation style for additional guidance.