parallel-web / parallel-agent-skills

Web search and page fetching for AI coding agents, through Parallel (opens in a new tab) — plus Agent Skills (opens in a new tab) for deep research, data enrichment, entity discovery, and web monitoring.

The plugin ships two things:

• A hosted MCP server (parallel-search). In Claude, it provides anonymous web_search, web_fetch, and web_decisions, plus web_research with Parallel sign-in. Authenticated calls are billed to your Parallel account. The Codex and Antigravity packages retain anonymous, rate-limited search and fetch.
• CLI-backed skills for deep research, enrichment, FindAll, and Monitor. These run through parallel-cli and need it installed, authenticated, and funded.

You can use the first without ever setting up the second.

Prerequisites

Claude Search, Fetch, and Decisions work without a Parallel account, subject to anonymous usage limits. Research requires a Parallel account with sufficient balance. No API key or local CLI is needed for MCP tools. Codex and Antigravity retain anonymous search and fetch with no account required.

The execution skills require parallel-cli (installed, authenticated, and funded). The parallel-cli-setup skill walks an agent through install, auth, balance, and skills install end-to-end — install the plugin/skills below, then run /parallel:parallel-cli-setup from your agent.

migrate-to-parallel updates an application’s own web-data integration. It uses the appropriate Parallel API or SDK and needs PARALLEL_API_KEY only for an explicitly authorized live smoke test.

Installation

Agent Skills

Use parallel-cli to install skills into Cursor, Cline, GitHub Copilot, and other compatible agents.

# Install all Parallel skills
parallel-cli skills install

Claude Code

Available as a Claude Code Plugin Marketplace (opens in a new tab).

/plugin marketplace add parallel-web/parallel-agent-skills
/plugin install parallel
# restart Claude Code before continuing!

After restarting, run /mcp to inspect the plugin-provided parallel-search server. Search, Fetch, and Decisions support anonymous use. Research requires Parallel sign-in; in clients that support tool-level authentication, requesting Research prompts you to connect your account. Once signed in, all authenticated MCP calls are billed at Parallel API rates (opens in a new tab).

Some Claude clients do not complete tool-level sign-in or expose Research before authentication. If that happens, use the connection’s sign-in control. If the server supplies an authenticated fallback URL, review the account and billing implications before switching to it. Do not silently change endpoints or add a duplicate server. See the setup skill for the explicit-login fallback.

Upgrading from v0.10.0 changes the Claude plugin endpoint to unified MCP at https://mcp.parallel.ai/mcp/claude. Update the plugin, fully exit Claude Code, and start a fresh session. Existing conversations can retain old tool arguments after /reload-plugins; if a call rejects model_name or session_id, start a fresh session and use the current schema. An old endpoint’s login may not carry over. If you previously added a separate Parallel connection, verify the new one works before disabling the duplicate.

For the CLI-backed skills (deep research, enrichment, FindAll, Monitor), install and authenticate the CLI:

# this will install/update CLI and authenticate if not done already
/parallel:parallel-cli-setup

OpenAI Codex

Install skills using the built-in skill installer (run inside Codex):

$skill-installer parallel-web/parallel-agent-skills

Then run the setup skill to install/auth the CLI:

/parallel:parallel-cli-setup

Google Antigravity

Install the standalone Search MCP plugin with the Antigravity CLI:

git clone https://github.com/parallel-web/parallel-agent-skills.git
cd parallel-agent-skills
agy plugin install ./plugins/antigravity/parallel

This package provides free web search and page fetching without a Parallel account, API key, or local Parallel CLI. Use the package directory above so Antigravity loads only the Search MCP connection. See the Antigravity package README (opens in a new tab) for project-local installation, validation, example prompts, and removal.

CDN and discovery

A human + machine-readable catalog is published at skills.parallel.ai (opens in a new tab).

Useful endpoints:

• skills.parallel.ai (opens in a new tab) — human-friendly catalog and install instructions
• skills.parallel.ai/index.json (opens in a new tab) — machine-readable skill index
• https://skills.parallel.ai/<skill>/SKILL.md — live raw skill file
• https://skills.parallel.ai/<skill>/manifest.json — file manifest + checksums
• https://skills.parallel.ai/<skill>/versions.json — release history for that skill
• https://skills.parallel.ai/archives/<skill>/<version>.zip — immutable GitHub Release archive via CDN redirect

Skills

Skills follow the Agent Skills (opens in a new tab) specification and double as Claude Code slash commands.

Examples

/parallel:parallel-web-search latest React 19 features
/parallel:parallel-search-evaluation compare search providers on production questions
/parallel:parallel-web-extract https://docs.parallel.ai
/parallel:parallel-deep-research competitive landscape of AI code assistants
/parallel:parallel-data-enrichment Apple, Microsoft, Google - get CEO names
/parallel:parallel-findall AI startups that raised Series A in 2026
/parallel:parallel-monitor track price changes for the iPhone 16 Pro
/parallel:parallel-memory retrieve past research about AI code assistants
/parallel:migrate-to-parallel migrate this app from Tavily to Parallel
/parallel:migrate-to-parallel migrate this app from Perplexity to Parallel
/parallel:migrate-to-parallel migrate this app from Firecrawl to Parallel
/parallel:parallel-cli-setup

Accounts, billing, and free access

Claude

The Claude plugin connects to https://mcp.parallel.ai/mcp/claude. Search, Fetch, and Decisions accept anonymous calls. Research requires OAuth and usable account credits. After sign-in, authenticated calls are billed under your account’s pricing and limits (opens in a new tab). Manage balance and usage at platform.parallel.ai (opens in a new tab).

MCP research returns a cited answer. The existing deep-research skill continues to run asynchronous Task API reports through the CLI; it is a separate workflow.

Codex and other anonymous Search MCP installations

The bundled parallel-search MCP server is anonymous by design: web_search and web_fetch work with no account, API key, or login, subject to free-tier rate limits. The free tier is intended for personal agents, hobby projects, and exploration.

For higher search limits, usage analytics, and production workloads, create an account at platform.parallel.ai (opens in a new tab), then add a separate authenticated Search MCP connection. The same service supports a Parallel API key on https://search.parallel.ai/mcp or OAuth through https://search.parallel.ai/mcp-oauth. Authenticated usage is billed under your account’s pricing and limits — see parallel.ai/pricing (opens in a new tab).

Keep the plugin-provided connection anonymous rather than editing the installed plugin. After the authenticated connection works, use /mcp to ensure only one Parallel Search connection is enabled; if both appear, toggle the bundled anonymous server off. See parallel-search-setup for the upgrade steps.

The CLI-backed skills are separate: deep research, enrichment, FindAll, and Monitor run through parallel-cli against your own funded Parallel account and are billed as ordinary API usage.

Privacy and terms

Parallel hosts the MCP service. Tool arguments sent to Parallel can include objectives, search queries, requested URLs, research questions, decision inputs, and optional session and model metadata. Parallel also receives standard connection and MCP client metadata. This data is handled under the Customer Terms (opens in a new tab) and Privacy Policy (opens in a new tab). The MCP server does not automatically receive the rest of your conversation or local files, but an agent can include context in tool arguments when it is relevant. Do not send secrets, credentials, or sensitive local content through MCP tools.

The CLI-backed skills run parallel-cli locally using credentials you have already configured. They send task inputs to Parallel’s API under the same terms.

Claude uses OAuth when account access is needed. Codex and Antigravity retain anonymous MCP connections. No bundled connection forwards CLI credentials. CLI-backed skills authenticate separately to Parallel through parallel-cli; the setup skill can check authentication status and start the CLI’s sign-in flow.

Migration examples describe authentication for the application being migrated. Provider credentials shown in scanner tests are synthetic source-code fixtures, not real keys or live requests. Neither those examples nor the test fixtures run automatically when the plugin loads.

Support and security

• Questions and bugs: open an issue (opens in a new tab) with your client, version, and error message (leave out API keys and other credentials), or see the documentation (opens in a new tab).
• Product support: contact Parallel at support@parallel.ai (opens in a new tab).
• Security concerns: report privately to support@parallel.ai (opens in a new tab) — see SECURITY.md (opens in a new tab). Please don’t file security reports as public issues.

Contributing

See MAINTAINERS.md (opens in a new tab) for maintainer workflows, release process, and dev setup.

Resources

• Documentation (opens in a new tab)
• API Platform (opens in a new tab)
• parallel-cli (opens in a new tab)
• Pricing (opens in a new tab)

Local Development

Claude Code:

git clone https://github.com/parallel-web/parallel-agent-skills.git
claude --plugin-dir /path/to/parallel-agent-skills
/parallel:parallel-cli-setup

Codex:

git clone https://github.com/parallel-web/parallel-agent-skills.git
cd parallel-agent-skills
codex
# Skills are auto-discovered via .agents/skills/

License

MIT