bug: get_run_details has returned a non-progressing zero-limit cursor since PR #3224

[lokicik]

Provide environment information

System:
  OS: macOS 26.5.2
  CPU: (10) arm64 Apple M4
  Memory: 1.82 GB / 16.00 GB
  Shell: 5.9 - /bin/zsh
Binaries:
  Node: 24.19.0
  npm: 11.7.0
  pnpm: 11.19.0
  bun: 1.3.14
Trigger.dev:
  CLI and SDK: 4.5.11
  MCP transport: stdio
  Flags: --dev-only --readonly
  Environment: dev

Describe the bug

The MCP get_run_details tool accepts maxTraceLines: 0 and reports a
successful empty page with nextCursor: "0". Following that cursor requests
the same page again, so pagination cannot advance.

I found this while testing Trigger.dev's MCP surface with an internal agent
compatibility system at Bottomless. The reproduction is model-free and uses a
dedicated benchmark dev project.

Expected behavior: reject a page size below one, or return a cursor that
advances.

Actual behavior:

Run Trace (lines 1-0 of 43)
More trace available.
nextCursor: "0"

To reproduce

1. Start the official Trigger.dev MCP for a project that has a completed dev
   run.
2. Call get_run_details for that run with maxTraceLines: 0.
3. Observe that the call succeeds, reports lines 1-0, and returns cursor
   "0" while saying more trace is available.
4. Call the tool again with cursor "0"; it returns the same page.

Reduced arguments:

{
  "projectRef": "<project-ref>",
  "environment": "dev",
  "runId": "<completed-run-id>",
  "maxTraceLines": 0
}

A control call with maxTraceLines: 1 returns lines 1-1 and cursor "1".

Additional information

I traced the behavior through Main history. The non-progressing cursor path was
introduced by PR #3224,
feat(cli): Expand and improve the MCP server and dev CLI command, merged on
March 17, 2026 at commit
dbbe9f77f953ad7a8306e48d7bd1107e3f8d3ba2.

Before that PR, maxTraceLines existed but get_run_details did not return a
cursor. PR #3224 added cursor pagination while leaving the zero value valid.
Its new readLinesPage function computed nextCursor as offset + limit, so
a zero limit returned the same cursor indefinitely.

The schema still accepted every integer when I rechecked Main on August 20,
2026 at 447471843cbe7759f64beb616d5db5f05696ea13, because maxTraceLines has
no minimum:

trigger.dev/packages/cli-v3/src/mcp/schemas.ts

Lines 146 to 160 in 4474718

	export const GetRunDetailsInput = CommonRunsInput.extend({
	maxTraceLines: z
	.number()
	.int()
	.describe(
	"The maximum number of trace lines per page. Defaults to 200. Use the `cursor` parameter to fetch subsequent pages."
	)
	.default(200),
	cursor: z
	.string()
	.describe(
	"Pagination cursor returned from a previous get_run_details call. Pass this to fetch the next page of trace events."
	)
	.optional(),
	});

The same cursor calculation also remains on current Main:

trigger.dev/packages/cli-v3/src/mcp/tools/runs.ts

Lines 35 to 52 in 4474718

	/** Read a page of lines from a file. Returns the lines and whether there are more. */
	function readLinesPage(
	filePath: string,
	offset: number,
	limit: number,
	totalLines: number
	): { lines: string[]; hasMore: boolean; nextCursor: string | null } {
	const content = fs.readFileSync(filePath, "utf-8");
	const allLines = content.split("\n");
	const pageLines = allLines.slice(offset, offset + limit);
	const end = offset + limit;
	const hasMore = end < totalLines;
	return {
	lines: pageLines,
	hasMore,
	nextCursor: hasMore ? String(end) : null,
	};

Suggested change: require a positive integer in GetRunDetailsInput, add a
defensive positive-limit check in readLinesPage, and cover zero/negative
limits plus a one-line positive case in tests.

The live check had zero invalid cases, unchanged fixture state, and verified
cleanup. Searches of existing issues and PRs found no matching report. This is
an MCP response-contract issue, not a security report.

We are testing a system at Bottomless that runs repeatable MCP contract and
workflow checks, verifies the observed behavior directly, and can replay the
same checks when an MCP changes. We are trying to learn whether MCP teams would
want checks like this on future pull requests or releases. Would that be useful
to your team?

If useful, I can look into a focused regression test and fix, then rerun the
same check.