MCP limits

MCP clients work best with bounded responses. Sleepwalker keeps list and result payloads compact and exposes drill-down tools for larger artifacts.

Tool Limits

Area	Limit
list_sleepwalker_tests	Maximum 50 tests.
get_sleepwalker_summaries	1 to 30 days, maximum 100 summaries.
get_sleepwalker_test_results	Maximum 5 runs.
get_sleepwalker_reports_by_url	Maximum 20 reports.
list_sleepwalker_visibility_runs	Maximum 100 runs.
list_sleepwalker_content_runs	Maximum 100 runs.
Page serialization chunks	max_chars is capped at 25,000.
Prompt response chunks	max_chars is capped at 25,000.
Content recommendation output	Compact recommendation sets by default.

See Limits for shared API/MCP limits.

Large Visibility Runs

Use one run creation call for batch work.

Good:

create_sleepwalker_visibility_run with 100 prompts x 4 platforms

Avoid:

400 separate MCP calls, one for each prompt/platform pair

Large runs should be read through status summaries, probe rows, and prompt-response drilldowns.

Polling

Poll persisted run status every 10 to 30 seconds.

Avoid tight loops. If a run is still queued or running, wait before polling again.

Response Size

Use:

• max_chars and offset for long serialized pages.
• get_sleepwalker_prompt_response for one prompt response at a time.
• include_result and include_results only when the user needs full details.

URL Safety

Sleepwalker rejects unsafe or unsupported URLs before execution.

Agents should provide normal public http or https URLs and avoid local, private, or credential-bearing URLs.