mcpField
W
web-research

web-research

github.com/pollinationsProductivity5.3k165 installsUpdated today
Repository

About web-research

Web Research is a shell-script-based research skill that queries web-search models for answers. It is intended for users who need sourced research, with options to select or compare models.

Key features

  1. 01Queries two models in parallel by default
  2. 02Select a model for a query
  3. 03Compare multiple models
  4. 04API key via environment variable or local file

Use cases

  1. 01Research a question using web search
  2. 02Fact-check a claim with sources
  3. 03Compare answers from several models
  4. 04Review a build plan with preliminary searches

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/pollinations/pollinations --skill web-research

After installing, type/web-researchin Claude Code to confirm it loaded.

Bundled filesscripts/web-research.sh

Details

From the project README

Web Research

Available Models

  • gemini-search — Google Gemini with web search grounding (default)
  • perplexity-fast — Perplexity AI, faster (default)
  • perplexity — Perplexity AI
  • nomnom — NomNom search model

By default, queries both gemini-search and perplexity-fast in parallel.

Requirements

  • curl

Authentication

Set an API key in an environment variable (preferred):

export POLLINATIONS_API_KEY="YOUR_KEY"

Or create a local .env file at .claude/skills/web-research/.env:

POLLINATIONS_API_KEY="YOUR_KEY"

If POLLINATIONS_API_KEY is not set, the script will prompt for a key (input hidden).

Quick usage

.claude/skills/web-research/scripts/web-research.sh "What is pollinations.ai?"

Choose a model:

.claude/skills/web-research/scripts/web-research.sh --model perplexity-fast "Fact-check this claim with sources"

Compare multiple models:

.claude/skills/web-research/scripts/web-research.sh --models gemini-search,perplexity-fast,nomnom "Compare answers"

Run multi-model in parallel:

.claude/skills/web-research/scripts/web-research.sh --models gemini-search,perplexity-fast --parallel "Compare answers"

Before reviewing a build plan

Run two searches first — "<vendor> <capability> <year>" and "<capability> MCP server" — and record the result. A first-party vendor launch in the last two quarters gets its own section ("why not the managed product"). It's the cheapest part of the review and can invalidate the whole plan, so do it before studying the design.

Numbers from PDFs or long documents

A summarizer given a leading prompt ("did X beat Y?") tends to echo the prompt, including invented quotes. Ask neutrally ("list every numeric result table") and, for any number you'll repeat, read the source text or table yourself.

Notes

  • Uses https://gen.pollinations.ai/v1/chat/completions
  • Sends Authorization: Bearer <key>

Comments

Markdown supported