Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Web Research for AI Agents

    1

    Real research instead of one search call: nine engines, real page reading, and a browser your agent can drive.

    $12

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Web Research for AI Agents

    Web Research for AI Agents

    Example session with this skill installed

    Compare the async HTTP clients we could use for a rate-limited public API
    client, and recommend one. Cite real sources and flag anything you could not
    verify.

    • Read your context and instructions
    • Compiled the web research

    Recommended: httpx with an explicit retry transport

    httpx is the closest fit for a rate-limited client because retry, backoff and
    rate limiting are things you add deliberately rather than behaviours you
    inherit.

    ClientAsyncRetries built inWhy it fits / does not
    httpxYesNo - you add a transportExplicit retry config, HTTP/2, sync + async
    aiohttpYesNoFast and mature, but its own idioms, heavier to mock
    requestsNoNoSync only - wrong shape for a rate-limited client
    urllib3PartialYesRetries exist but the high-level API is awkward

    The detail that decides it

    httpx does not retry by default. That is a feature here, not a gap: on a
    rate-limited API you want to control the backoff yourself, honour
    Retry-After, and cap total attempts. A client that silently retries hides the
    moment you are actually being throttled.

    What I could not verify

    Whether your API returns Retry-After in seconds or as an HTTP date. It
    changes the backoff implementation, so check one real 429 response before
    writing the transport.

    Next steps

    1. Pin the version in your lockfile.
    2. Add a retry transport with jitter and a hard attempt cap.
    3. Log every 429 with its headers so throttling is visible.

    Sources: github.com/encode/httpx (client and transport documentation),
    pypi.org/project/httpx

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Your agent already knows how to search. What it usually does is call one search tool, trust the snippets, and write a confident paragraph from them. That is how you get citations that do not support the claim.

    This pack is the workflow layer for the websearch MCP server: the ordered sequences that turn a search call into actual research, plus what to do when a page fights back.

    What you get

    • 25 worked scenarios. Each names the tools, the arguments, the stop condition, and the mistake people make. Real tasks: verifying a disputed claim, comparing five products, extracting a table, finding who said a quote, checking a price today, handling a JS-rendered page.
    • A failure playbook. Thirteen failure modes with Symptom / Cause / Do this / Do not do this. Empty results, cookie walls, JS shells, 429s, three tools that are silently unavailable, and the one case where the honest answer is "not found".
    • Token budget. When depth=1 is enough, when it is not, how to cap page size, and when synthesis costs more than it is worth.
    • Five client configs. Copy-paste setup for Claude Desktop, Cursor, Cline, Continue and Zed.

    Why this and not the repo

    The server is free, MIT, and on PyPI. Run pip install skdev-websearch-mcp and you have all eleven tools. This pack is the part the repo does not contain: which tool to call, in what order, with what arguments, and what to do when the result is empty or wrong.

    Requires

    • The websearch MCP server (pip install skdev-websearch-mcp)
    • No API keys. The keyless lanes work out of the box.
    • Optional: a Chromium-based browser with --remote-debugging-port for Google AI Mode, browser control, and bot-walled pages.

    Requires the websearch MCP server. Browser-based lanes (browser_action, CDP) function on local machines only and will not work on headless server deployments.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 23 days ago

    Requires

    Archive
    Websearch MCP
    Python 3.11+
    Skdev Websearch MCP

    Permissions

    Browser
    Network Access
    Environment Variables

    Allowed Hosts

    s
    web.archive.org

    Network access is needed because the server queries live search and page APIs. Browser access is optional and only used for Google AI Mode, browser control, and bot-walled pages. Environment variables are optional keys that upgrade existing lanes.

    Listed23 days ago

    What's inside

    Frequently Asked Questions

    This skill provides standardized logic, tool-call sequences, and recovery playbooks that prevent the agent from giving up when it hits a paywall, a JS-only site, or a bot-protection screen.

    It is compatible with any AI client or IDE (such as Claude Desktop or Cursor) that supports the Model Context Protocol (MCP) and has the 'skdev-websearch-mcp' server installed.

    You receive the complete production-grade research playbook, including diagnostic routines, the three-tier escalation ladder, and specific failure recovery sequences for various web obstacles.

    Yes, the 'doctor()' diagnostic routine included in the pack automatically identifies which search and browser lanes are active to ensure the agent doesn't attempt unsupported actions in restricted environments.

    Unlike generic prompting, this pack uses strict exit conditions and a tiered ladder approach that prioritizes lightweight fetches before escalating to resource-heavy browser actions, significantly reducing unnecessary token spend.

    The skill is designed to be plug-and-play with the standard websearch MCP server; you simply import the workflow definitions into your MCP-enabled client to begin using the standardized patterns.

    New in LLM Ops