get_agent_run_trace
Use this tool with agents built with eve.
Read one page of an eve Agent Run trace, including turns, messages, reasoning, tool calls, token usage, and tool input or output when available.
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
teamId | string | Yes | - | The team ID or slug. Use the team that owns the resource. |
projectId | string | Yes | - | The project ID or slug. Find the project ID in .vercel/project.json or use list_projects. |
runId | string | Yes | - | The Agent Run ID to inspect. |
environment | string | No | - | Agent run environment, usually "production" or "preview". Defaults to "production". |
period | string | No | - | Preset time range. Ignored when both from and to are provided. Defaults to the dashboard endpoint default. Allowed values: "5m", "15m", "1h", "6h", "12h", "1d", "3d", "7d", "14d", "30d", "90d". |
from | string | No | - | Start time as ISO 8601, Unix seconds, Unix milliseconds, or a relative duration like "12h". Must be used with to. |
to | string | No | - | End time as ISO 8601, Unix seconds, Unix milliseconds, a relative duration like "1h", or "now". Must be used with from. |
maxFieldLength | number | No | - | Maximum length for individual string fields in the returned trace. Defaults to 8000. Use 0 to disable truncation. |
traceStreamId | string | No | - | Copy streamId from the trace payload's later or earlier object. Requires exactly one pagination cursor. |
traceAfterCursor | string | No | - | Copy cursor from later to fetch the next page, with its traceStreamId. |
traceBeforeChunk | integer | No | - | Copy beforeChunk from earlier to fetch the previous page, with its traceStreamId. |
The response contains a trace object with runId, framework, optional metadata, and the framework's trace payload. Current eve runs use trace.eve. Older payloads can use trace.ash. Inspect the payload before reading its continuation fields.
| Payload field | Meaning |
|---|---|
turns | The turns returned on this page. Turns can include messages, reasoning, tool calls, and token usage. |
later | A forward continuation containing streamId and cursor. Null or absent when there are no more currently persisted chunks. |
earlier | A backward continuation containing streamId and beforeChunk. Null or absent when the page reaches the stream's start. |
truncated | Optional metadata with droppedSide: oldest, newest, or both. Indicates which part of a bounded stream read is missing. |
unavailableReason, unavailableCode | Optional explanation when trace content is unavailable. An empty or unavailable trace does not imply the run had no activity. |
To read the next page, keep the same team, project, run, and time window. Copy later.streamId into traceStreamId and later.cursor into traceAfterCursor. For example, if later contains streamId: "strm_123" and cursor: "opaque_cursor", call the tool with:
{
"teamId": "team_123",
"projectId": "prj_123",
"runId": "wrun_123",
"traceStreamId": "strm_123",
"traceAfterCursor": "opaque_cursor"
}Continue until later is null or absent. To read backwards, use earlier.streamId as traceStreamId and earlier.beforeChunk as traceBeforeChunk. Supply exactly one direction with traceStreamId.
Pages can overlap. Deduplicate turns by ID and keep the more complete turn. String fields longer than maxFieldLength end with a truncation marker. Use maxFieldLength: 0 to preserve full strings. The MCP server preserves continuation fields even when it truncates other fields.
Sample prompt: "Show me the tool calls and messages from Agent Run wrun_123"
Was this helpful?