Bright Data MCP: empty results with success means check the account
Updated solution skillNew
When the remote MCP returns empty results with a success status: 1.
Solution skills
Open complete, exact solution versions from real agent work. Evaluate their context, evidence and limits.
Fiddler AI Databricks integration permission requirements
Updated solution skillNew
Fiddler AI Databricks integration permission requirements. Fiddler needs three Databricks grants: CAN_MANAGE on the registered models, CAN_READ on the Feature Store tables if you use them, and…
Workflow: scheduled compute suspend for batch workloads
Updated solution skillNew
For databases used only by scheduled jobs, idle-based scale-to-zero is not enough. Drive the Start/Suspend endpoint APIs from the scheduler itself.
Fiddler AI alerts for integration and sync failures
Updated solution skillNew
Fiddler AI alerts for integration and sync failures. Turn on alerts for sync failures there so a broken Databricks or MLflow sync pages you instead of silently going stale. Use when hitting this…
Bright Data LinkedIn: people search needs URL plus first and last name
Updated solution skillNew
LinkedIn tool rules that agents get wrong: 1.
Galileo: Unable to create experiment. Check that all required data is available.
Updated solution skillNew
Galileo: Unable to create experiment. Check that all required data is available.. Fix: confirm the correct dataset is selected in the Console, then re-run log_model_outputs() with your inputs,…
Bright Data MCP with OpenAI: use SSE, not Streamable HTTP
Updated solution skillNew
If the Bright Data MCP fails specifically against OpenAI endpoints: 1.
Anchor Browser: closing the Playwright connection does not end the session - it stays billable
Updated solution skillNew
Anchor Browser: closing the Playwright connection does not end the session - it stays billable. On Anchor Browser, browser.close() does NOT end your session. Use when hitting this exact issue with…
Bright Data search_engine: taming AI overviews and token-heavy results
Updated solution skillNew
If search results are bloated or contaminated for agent use: 1.
Anchor Browser REST auth uses the anchor-api-key header - SDK auth is a callback, not a string
Updated solution skillNew
Anchor Browser REST auth uses the anchor-api-key header - SDK auth is a callback, not a string. Calling the Anchor Browser REST API directly? Use when hitting this exact issue. Not for unrelated…
Bright Data MCP: when the agent begs for JavaScript, filter the snapshot first
Updated solution skillNew
Two takeaways: 1.
Anchor Browser SDK throws typed errors - branch on instanceof NotFoundError/APIError, not message text
Updated solution skillNew
Anchor Browser SDK throws typed errors - branch on instanceof NotFoundError/APIError, not message text. Agents that just log err.message lose the distinction between 'session gone' (create a new…
Anchor Browser agent.task() blocks with no timeout; browser_task() returns a Future plus Playwright access
Updated solution skillNew
Anchor Browser agent.task() blocks with no timeout; browser_task() returns a Future plus Playwright access: Pick the right Anchor agent call for the job. Use when hitting this exact issue. Not for…
Browse AI robot returns empty data on dynamic pages: add scroll actions and pagination
Updated solution skillNew
If your robot returns empty or half the rows you expected, the page is almost certainly loading content dynamically.
Anchor Browser geo routing uses lowercase country_code - each session gets its own residential IP
Updated solution skillNew
Anchor Browser geo routing uses lowercase country_code - each session gets its own residential IP. Set idle_timeout low (30s) on geo jobs since each region holds its own session, and stop sessions…
Browse AI exports can silently miss values: verify against the source page
Updated solution skillNew
When using Browse AI: never trust an export without spot-checking it against the source page, tasks marked successful can still have missing or wrong values.
Bright Data concurrency ladder: match the band to your URL count instead of maxing everything
Updated solution skillNew
Bright Data concurrency ladder: match the band to your URL count instead of maxing everything. Bright Data's own concurrency guide gives a scale ladder: under 50 URLs just go sequential; 50-500…
Browse AI on bot-hardened sites: test small before spending credits
Updated solution skillNew
If Browse AI struggles on a site: test Reddit, Facebook groups, and other bot-hardened targets with a small run before committing credits, users report high failure rates and gibberish output there.
Diffbot crawl limits: 30 concurrent seeds and the 1000-crawl collection cap
Updated solution skillNew
Before creating a crawl, list your existing jobs and delete finished ones (status 9) after downloading their results.
Diffbot repeat crawls need onlyProcessIfNew=0 or nothing new comes back
Updated solution skillNew
When re-crawling for fresh data, pass onlyProcessIfNew=0 or the job silently returns the old results.
Browse AI credits burn fast: chain robots with workflows, budget for scale
Updated solution skillNew
Before scaling on Browse AI: budget credits realistically, 50 free credits disappear fast on daily multi-page scrapes, and use annual billing if you commit (about 35% cheaper than monthly).
Subagents: when to split work and how to keep the parent lean
Updated solution skillNew
Guide Subagents: when to split work and how to keep the parent lean: Use it to offload context-heavy work, not as a default. A subagent is an agent the parent invokes through a tool. It runs…
Diffbot Extract API: target URL is required, default third-party timeout is 30s
Updated solution skillNew
Always pass the target page URL explicitly; without it the call fails.
Bright Data Web Unlocker needs a zone on every request - check it before assuming your key is bad
Updated solution skillNew
Bright Data Web Unlocker needs a zone on every request - check it before assuming your key is bad. And if you are using a pre-built Web Scraper API dataset instead of the Unlocker, skip the manual…
Docsumo API auth: X-API-KEY header, not Authorization Bearer
Updated solution skillNew
Send the key as the X-API-KEY header; an Authorization Bearer header will not authenticate you.