Bright Data MCP Amazon calls return 10k+ tokens per product — can I filter fields?
Using brightdata-mcp inside Claude, a single web_data_amazon_product call returns roughly 10k tokens, and an Amazon search page can return 50k–100k tokens, exhausting the context window when analyzing 10+ products. Is there a way to request only essential fields like title, price, rating, and review count? The issue was reported in July 2025 against the current MCP implementation, which returns full product data with no filtering options.
Originally asked on GitHub: https://github.com/brightdata/brightdata-mcp/issues/40
4 answers
Status: this issue is closed as completed, and a maintainer said the work is handled in PR #60 — but the thread does not confirm that the PR was merged or name a release version, so check the repository before relying on filtering parameters.
While you wait, there is no built-in way to filter fields in the version discussed here. The reporter's workaround was to limit analysis to 3–5 products or extract data manually. A related token-efficiency change was also announced: Rapid vs. Advanced modes, where Rapid mode would default to exposing only two tools (Search Engine and Scrape as Markdown) and setting
ADVANCED_MODEto true would expose all 60+ tools.The maintainer explained the design constraint: exposing a list of all available fields to the LLM could itself blow the context window —
tools/listalready consumes around 17K tokens with 60 tools — so they preferred trimming redundant output fields over offering a full field list. In response to a user suggestion, limiting the number of returned items per field (e.g., only 2 reviews, 2 image URLs) was acknowledged as something to consider.A community proposal to add offset/pagination parameters to every tool was ruled out: per the MCP 2025-03-26 spec, pagination is supported only for certain operations that do not apply to these tools.
Separately, one user reported that when the Amazon product reviews tool fails, the MCP may not report the failure and the call waits until timeout; that was being debugged by the scraper team at the time of the thread.
Did this work for you?What a GitHub user tried on GitHub (Outcome unclear)
The reporter's stated workaround was to limit analysis to 3–5 products or extract data manually to avoid exhausting the conversation context; the thread does not report how well this worked.
Source: https://github.com/brightdata/brightdata-mcp/issues/40 (opens external site)
Did this work for you?What @fahimanwer tried on GitHub (Outcome unclear)
A commenter suggested adding an offset parameter to every tool so large results could be fetched in parts, citing the MCP 2025-03-26 pagination spec. The maintainer replied that pagination is only supported for certain operations that are not relevant to these tools.
Source: https://github.com/brightdata/brightdata-mcp/issues/40#issuecomment-3193479924 (opens external site)
Did this work for you?What @fahimanwer tried on GitHub (Outcome unclear)
A user proposed adding an option to limit the number of outputs per field, e.g., only 2 reviews or 2 image URLs per product. A maintainer acknowledged the idea and said they would consider implementing it in the next release.
Source: https://github.com/brightdata/brightdata-mcp/issues/40#issuecomment-3172152858 (opens external site)
Did this work for you?
Related questions
- Does brightdata-mcp support password authentication instead of an account-wide API token?
- Bright Data MCP times out on LinkedIn scrape: how to avoid the 1-minute /scrape timeout?
- Why do Bright Data MCP Browser API tools fail only on some target sites?
- Bright Data MCP browser tools hang 5 min in Docker/Render but work locally
- How to run brightdata-mcp in Docker without the initialize timeout in Claude Desktop?