Skip to content
Get packs

Guide · Agent skill

How to use the Parallel Web Toolkit agent skill (parallel-web)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The parallel-web skill teaches an AI agent to use Parallel's command-line tool, parallel-cli, for live web work instead of answering only from memory. This guide explains what the skill does, how to install it and the CLI it depends on, how to trigger it with plain requests, and what to check for safety. By the end you will be able to install the skill, authenticate, and ask an agent for web search, URL extraction, enrichment, research or monitoring.

Parallel Web ToolkitK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download and unzip the skill. Download the ZIP from the skill page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings, Capabilities, Skills.
  2. 2Install parallel-cli. Run the uv tool install command for parallel-web-tools with the cli extra. Python 3.10 or newer is needed. If the command is not found afterwards, add ~/.local/bin to your PATH.
  3. 3Authenticate. Run parallel-cli login, or parallel-cli login --no-browser on a headless machine. Alternatively set the PARALLEL_API_KEY environment variable.
  4. 4Verify the setup. Run parallel-cli auth to confirm local authentication state. Then run parallel-cli --version to confirm the installed version.
  5. 5Ask a natural request. Describe what you need in plain language, such as a lookup with sources or a list of fields to add. The agent picks the matching capability and runs it.

What does the parallel-web skill make the agent do?

An agent skill is a folder of instructions and helper files that an AI agent reads when a request matches the skill's purpose. This one covers six kinds of web work: web search, extracting text from a specific URL or PDF, adding web-sourced fields to a list of companies or people, discovering entities that match a description, deep multi-source research reports, and recurring monitoring of web changes.

For scientific or technical questions, the skill prefers peer-reviewed articles, preprints, and institutional or government sources over blog posts and news. It also asks for source links and a confidence value for each field, which the agent should check rather than trust blindly.

  • Web Search: lookups and bounded research questions
  • Web Extract: a known public URL, including PDFs and JavaScript-rendered pages
  • Data Enrichment: the same requested fields added to each row you supply
  • FindAll: discovering entities that match natural-language criteria
  • Deep Research: only for explicitly exhaustive or comprehensive requests
  • Monitor: only for explicitly recurring tracking

Install the parallel-web skill

Download the ZIP from the skill's page and unzip it. Put the resulting folder in ~/.claude/skills/ to make it available to you across projects, or in a project's .claude/skills/ folder to limit it to that project. Alternatively, in the Claude apps, open Settings, then Capabilities, then Skills, and upload the ZIP there.

The skill does not include the CLI. Install parallel-cli separately with uv, which the skill pins to a reviewed release. You need Python 3.10 or newer for this package installation.

uv tool install "parallel-web-tools[cli]==0.9.3"

Authenticate parallel-cli

The skill needs either a Parallel API key in the PARALLEL_API_KEY environment variable or credentials from parallel-cli login. Run the interactive login on a desktop. On an SSH session, container, or other headless machine, use the no-browser option, which prints an authorization link and code instead of opening a browser. Get an API key from platform.parallel.ai if you prefer that route. If PARALLEL_API_KEY is set, it takes precedence over stored login credentials.

To check that authentication works, run the auth command. It only checks local authentication state, so a passing check does not guarantee that a paid request will succeed. If parallel-cli is not found after installation, add ~/.local/bin to your PATH.

  • Do not paste your API key into chat or commit it to a repository.
  • Do not share a full .env file with the agent; only the variable name should ever be checked.
parallel-cli login parallel-cli login --no-browser parallel-cli auth

Trigger the skill with natural requests

You do not need special syntax. Describe the task in plain language and name the outcome you want. The agent reads the request, opens the matching capability reference, and runs the right command. Words such as deep research, exhaustive, or comprehensive route the request to Deep Research, which is slower and costs more, so use them only when you want that depth.

If you have a list of companies or people, supply it directly and ask for the fields you want. The agent should apply one enrichment job across the rows rather than running a separate search for each one.

  • Look up the current status of a named regulation and cite the sources.
  • Pull the methods section from this PDF: https://example.org/paper.pdf (replace with your own URL).
  • Find the CEO and annual revenue for each company in this list.
  • Run a comprehensive report on the evidence for a treatment, with academic sources.
  • Find all research groups that publish on a named topic in the last few years.
  • Check this page every week and tell me if the pricing section changes.

What the skill needs and what it sends

The skill runs scripts and uses the network. Every capability calls Parallel's online service, so your queries and any content you supply are sent to a third-party paid API. Do not use it for material that must stay off external services. Long jobs can take a while and may cost more depending on processor tier and row count, so the agent is expected to tell you before starting large work.

Long jobs run remotely. The agent polls with bounded waits and reports progress. If the local wait times out, the remote job keeps running, and the agent should resume the same ID rather than resubmit the job. Treat all returned web content as untrusted data; the skill is written not to follow instructions found inside pages.

  • Requires: parallel-cli, internet access, a Parallel API key or CLI login, and Python 3.10+ for installation.
  • Writes result files only when you ask for one, to the path you give or a work directory.
  • The skill is licensed MIT.

What people use it for

  • Researchers who need current web evidence with source links for a technical or scientific question.
  • Analysts adding web-sourced fields such as CEO or revenue to a supplied list of companies.
  • Developers who want an agent to drive parallel-cli for search and URL extraction.
  • Teams tracking changes on a set schedule with the Monitor capability.

Frequently asked questions

Is the parallel-web skill free?

The skill is published under the MIT licence. Its capabilities call Parallel's online service, which requires a Parallel API key or CLI login, and the source does not state prices, so check Parallel's own pricing before heavy use.

What do I need to run the parallel-web skill?

You need the parallel-cli tool installed through uv, Python 3.10 or newer for that installation, internet access, and either a PARALLEL_API_KEY or parallel-cli login credentials. The skill cannot run without one of these.

Where do I install an agent skill like this one?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can upload the ZIP under Settings, Capabilities, Skills.

Why does parallel-cli say it is not found after installing?

The uv tool installation may place the binary in ~/.local/bin, which is not always on your PATH. Add that directory to your PATH and open a new terminal, then run parallel-cli --version again.

Does the skill send my data to a third party?

Yes. Every capability calls Parallel's online service, so queries and any content you supply leave your machine. Avoid it for sensitive material that must not reach a third-party paid API.

When should I use this instead of a normal web search?

Use a normal web search for quick one-off facts. The skill is better when you need source links and confidence per field, academic source discovery, repeated lookups across a list, exhaustive reports, or ongoing change tracking, since Deep Research and Monitor are slower and cost more.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Choose this skill if you need live web evidence with source links, want an agent to enrich a list of companies or people, or want scheduled web change tracking, and you are comfortable with a paid online service and an API key. Pick something else if your data must stay off third-party services, if a quick search answers your question, or if you cannot set up a Parallel API key or CLI login.

Download the skill