Skip to main content
RapidDev - Software Development Agency

How to Install and Use Tavily Web Search in OpenClaw

To use Tavily Web Search in OpenClaw, run `clawhub install tavily-web-search`, configure your TAVILY_API_KEY from app.tavily.com, and start querying. Unlike Tavily AI Search, this skill returns raw, unranked web results without AI post-processing — making it significantly faster and cheaper per query, ideal for bulk data collection and pipelines where you process results yourself.

What you'll learn

  • How to install tavily-web-search alongside (or instead of) tavily-ai-search using one API key
  • The specific performance and cost advantages of raw search over AI-enhanced search
  • How to write OpenClaw prompts that benefit from raw, unfiltered result sets
  • How to configure result count and domain filters for bulk data collection
  • When to route queries to tavily-web-search versus tavily-ai-search within the same OpenClaw instance
Book a free consultation
4.9Clutch rating
600+Happy partners
17+Countries served
190+Team members
Beginner12 min read5 minutesSearch & ScrapingMarch 2026RapidDev Engineering Team
TL;DR

To use Tavily Web Search in OpenClaw, run `clawhub install tavily-web-search`, configure your TAVILY_API_KEY from app.tavily.com, and start querying. Unlike Tavily AI Search, this skill returns raw, unranked web results without AI post-processing — making it significantly faster and cheaper per query, ideal for bulk data collection and pipelines where you process results yourself.

Raw Web Results at Speed: When to Choose Tavily Web Search Over Tavily AI Search

Most OpenClaw users start with Tavily AI Search because the AI-enhanced results are immediately useful — you get a synthesized answer plus ranked sources without any extra configuration. But Tavily Web Search exists for a different job: scenarios where you want the raw search results and plan to process, filter, or combine them yourself inside OpenClaw or downstream in a pipeline.

The practical difference is speed and cost. Tavily AI Search adds an AI post-processing step that improves relevance but takes 1-3 additional seconds per query and consumes more API credits. Tavily Web Search skips that layer entirely. For a single research query, this difference is imperceptible. For an automated workflow that runs 50 searches per hour, the difference in API credit consumption becomes significant — and Tavily Web Search's lower per-query cost means you can do much more on the same free or paid tier.

Another important use case is when you explicitly do not want AI ranking to filter your results. Tavily AI Search's ranking algorithm makes judgment calls about relevance that work well most of the time but can suppress minority viewpoints, niche sources, or highly specialized technical content that a broader audience might rank lower. If your workflow needs the full unfiltered result set — for example, a competitive monitoring pipeline that must capture every mention regardless of perceived quality — Tavily Web Search gives you that raw coverage. Since both skills share the same TAVILY_API_KEY, installing both and routing different query types to each is a common and cost-effective pattern.

Integration method

ClawHub Skill

Tavily Web Search installs as a ClawHub skill and shares the same TAVILY_API_KEY as the Tavily AI Search skill — you can have both installed simultaneously using one API key. The key architectural difference is that this skill skips Tavily's AI post-processing layer, which reduces latency and API credit consumption. Responses arrive faster and include raw extracted content from matched pages without relevance scoring or synthesized answers.

Prerequisites

  • OpenClaw installed and running on your system
  • ClawHub CLI available in your terminal
  • A Tavily API key from app.tavily.com — same key works for both tavily-ai-search and tavily-web-search
  • If you already have tavily-ai-search installed, no new API key is needed — just install this additional skill

Step-by-step guide

1

Install the Tavily Web Search Skill

Run the ClawHub install command for the raw search variant. If you already have tavily-ai-search installed, this second skill installs alongside it without conflict — both use the same TAVILY_API_KEY stored in your OpenClaw config. You do not need to duplicate or create a second API key. The skill package is separate from tavily-ai-search because it uses a different Tavily API endpoint and response handling logic. Installing both gives you access to both search modes within the same OpenClaw instance, and you can direct OpenClaw to use one or the other by specifying in your prompt whether you want a 'quick search' (web search) or a 'research search' (AI search). After running the install, verify the skill appears in your installed skills list. You should see both `tavily-ai-search` and `tavily-web-search` listed if you have both installed.

terminal
1# Install the raw web search skill
2clawhub install tavily-web-search
3
4# If you also want the AI-enhanced version, install both
5clawhub install tavily-ai-search
6
7# Verify both are installed
8clawhub list | grep tavily

Pro tip: You can install multiple search skills and they will all coexist. OpenClaw routes to the appropriate skill based on context and your prompt phrasing. Both tavily skills share the same TAVILY_API_KEY — no need to configure it twice.

Expected result: Both tavily-web-search and tavily-ai-search appear in `clawhub list` output. Installation confirmation shows the skill version and required config key (TAVILY_API_KEY).

2

Get Your Tavily API Key (Skip If Already Done)

If you installed tavily-ai-search previously, your TAVILY_API_KEY is already configured and you can skip to Step 3. Both Tavily skills share a single API key. If this is your first Tavily skill, go to app.tavily.com and sign up for a free account. Navigate to the API Keys section in the dashboard and click 'Create API Key'. Name it something like 'OpenClaw' so you can identify it later. Copy the generated key — it starts with `tvly-`. Tavily's free tier grants 1,000 API calls per month. Tavily Web Search queries are cheaper in credit terms than Tavily AI Search queries because they skip the AI post-processing step. This means your 1,000 monthly free requests go further when you use this skill for bulk or high-frequency queries. Note that the same API key appears in both the Tavily Web Search and Tavily AI Search dashboards — there is only one credential to manage for both skills.

Pro tip: Check whether TAVILY_API_KEY is already set before going to the Tavily dashboard: run `clawhub config get TAVILY_API_KEY`. If it returns a value, you are ready to proceed to Step 3.

Expected result: You have a Tavily API key (tvly-XXXX) that will be used for both tavily-web-search and tavily-ai-search if both are installed.

3

Configure the API Key

Add your Tavily API key to OpenClaw's configuration. If you set this key for tavily-ai-search previously, it is already available to tavily-web-search — no additional steps needed. For a fresh installation, set the key using the clawhub config command or by editing the .env file directly in your OpenClaw configuration directory. The key name must be exactly `TAVILY_API_KEY` with this exact capitalization — the skill reads this specific environment variable name. After setting the key, reload the OpenClaw configuration to make it active in the current session.

terminal
1# Set the API key (same key as tavily-ai-search)
2clawhub config set TAVILY_API_KEY tvly-your-api-key-here
3
4# Or add to ~/.openclaw/.env directly
5echo 'TAVILY_API_KEY=tvly-your-api-key-here' >> ~/.openclaw/.env
6
7# Reload configuration
8clawhub reload
9
10# Confirm it is set
11clawhub config get TAVILY_API_KEY

Pro tip: The variable name TAVILY_API_KEY (uppercase, underscore) must match exactly. A common mistake is using TAVILY-API-KEY (with hyphens) or tavily_api_key (lowercase), which will cause authentication failures.

Expected result: `clawhub config get TAVILY_API_KEY` returns the key value (masked). OpenClaw confirms tavily-web-search is active and authenticated.

4

Test with a Raw Search Query in OpenClaw Chat

Open OpenClaw chat and run a test query. Because Tavily Web Search returns raw results rather than synthesized answers, your test prompt should ask for a list of results rather than a summary or explanation. The response from Tavily Web Search includes: a list of results with titles, URLs, and extracted page content snippets. Unlike Tavily AI Search, you will not see a synthesized paragraph at the top — just the ranked list of raw results. This is the expected behavior and the feature, not a bug. To explicitly direct OpenClaw to use the raw search skill rather than the AI search skill, include phrasing like 'get raw results', 'list all sources without summarizing', or 'fast search' in your prompt. This helps OpenClaw's router prioritize speed over synthesis.

OpenClaw Prompt

Do a fast web search for 'open source LLM releases in 2026' and give me a raw list of all results with their titles and URLs. Don't summarize — I want the full result set.

Paste this in OpenClaw chat

Pro tip: If you want to see which skill OpenClaw is using for a given query, enable verbose logging in your OpenClaw config (`log_level: debug`). This shows skill selection decisions in the terminal output.

Expected result: OpenClaw returns a structured list of 5-10 search results with titles, URLs, and content excerpts — without a synthesized summary paragraph. Response arrives in 1-3 seconds.

5

Configure for Bulk and High-Frequency Use

For pipelines or monitoring workflows that run many queries, fine-tune the skill configuration to balance coverage against API credit consumption. The key settings for high-frequency use are `max_results` (how many results per query) and `days` (recency filter to avoid re-fetching old content). RapidDev can help teams design OpenClaw pipelines that combine Tavily Web Search with downstream processing skills for competitive intelligence, content monitoring, and research automation. The configuration shown here covers the most common production tuning parameters. For workflows running more than 500 queries per month, plan to upgrade to a Tavily paid tier — the Research plan ($35/month for 10,000 requests) is the most popular choice for teams using OpenClaw for automated monitoring.

OpenClaw Prompt

Run a search for 'AI agent frameworks GitHub releases' and return all results from the last 7 days. List just the titles and URLs, sorted by publication date.

Paste this in OpenClaw chat

tavily-web-search.yaml
1# ~/.openclaw/skills/tavily-web-search.yaml
2tavily-web-search:
3 max_results: 10 # default: 5, increase for broader coverage
4 days: 7 # only fetch content from last N days
5 include_answer: false # skip synthesis raw results only (default for this skill)
6 include_raw_content: false # set true to get full page text (large responses)
7 include_domains: [] # optional: restrict to specific domains
8 exclude_domains: # optional: filter out low-quality sources
9 - pinterest.com
10 - quora.com

Pro tip: Setting `exclude_domains` for known low-signal sites like content farms and social aggregators improves result quality without needing AI post-processing. This is a free way to improve raw results before they reach OpenClaw.

Expected result: Searches return up to 10 results per query filtered to the last 7 days, with excluded domains removed. Queries complete in under 2 seconds on average.

Common use cases

Bulk URL Collection for Data Pipelines

Run many parallel searches to collect URLs and page content across a topic, then feed those results into a secondary processing step — deduplication, summarization, or structured data extraction. Raw results are better here because you want all matches, not a pre-filtered subset.

OpenClaw Prompt

Search the web for all news articles mentioning 'Series A funding' and 'AI startup' published in the last 24 hours. Return the full list of URLs and article titles without filtering or summarizing.

Copy this prompt to try it in OpenClaw

High-Frequency Monitoring Without Credit Burn

Set up an OpenClaw workflow that checks a topic every hour for new content. At this frequency, using Tavily AI Search would burn through API credits quickly. Tavily Web Search's lower credit cost makes hourly monitoring sustainable on the free tier.

OpenClaw Prompt

Search for any new announcements from OpenAI published in the last hour. Return the raw list of results including titles, URLs, and publication times.

Copy this prompt to try it in OpenClaw

Feeding Raw Search Results into a Summarize Skill

Use Tavily Web Search to collect a broad set of results, then pass those results to a separate OpenClaw summarization skill to generate the synthesized output. This two-step approach gives you more control over which content gets summarized compared to relying on Tavily AI Search's built-in synthesis.

OpenClaw Prompt

Search for 'Rust programming language performance benchmarks 2026' and return the raw page content from the top 8 results. I'll process the results myself.

Copy this prompt to try it in OpenClaw

Troubleshooting

tavily-web-search returns '401 Unauthorized' or 'Invalid API key'

Cause: TAVILY_API_KEY is not set, was set under a different variable name, or the key was entered incorrectly with extra whitespace or truncation.

Solution: Run `clawhub config get TAVILY_API_KEY` to check the current value. If empty, set it with `clawhub config set TAVILY_API_KEY tvly-your-key`. If set but still failing, generate a new key in the Tavily dashboard — keys can be revoked or expired.

typescript
1# Verify and reset
2clawhub config get TAVILY_API_KEY
3clawhub config set TAVILY_API_KEY tvly-XXXXXXXXXXXXXXXXXXXX
4clawhub reload

clawhub install tavily-web-search fails with a version conflict error

Cause: An older version of the skill or a conflicting dependency is already installed. This can happen if a previous install was interrupted or if the ClawHub package cache is stale.

Solution: Force a clean reinstall using the `--force` flag. This removes the existing package state and downloads a fresh copy from the ClawHub registry.

typescript
1clawhub uninstall tavily-web-search
2clawhub install tavily-web-search --force

Results from tavily-web-search look identical to tavily-ai-search results

Cause: OpenClaw may be routing both query types through the same skill, or the skill selection logic is defaulting to one skill regardless of prompt. This is a routing configuration issue, not an API issue.

Solution: Explicitly test the skill by name in your prompt. Check that both skills are installed and active in `clawhub list`. If OpenClaw's auto-routing is not distinguishing between them, you may need to configure explicit skill routing in your OpenClaw agent settings.

typescript
1# List installed skills to verify both are present
2clawhub list | grep tavily
3
4# Force-invoke a specific skill in your prompt
5# Use: 'Use tavily-web-search to find...'

OpenClaw is hitting Tavily's monthly rate limit before the month ends

Cause: High-frequency automated queries or bulk data collection workflows are consuming the 1,000 monthly free requests faster than expected.

Solution: Check your Tavily dashboard for usage statistics to see which time periods have the highest consumption. Add a `days` filter to avoid re-fetching content you have already seen. Consider upgrading to a paid Tavily tier, or shift some query traffic to Brave Search (which has its own separate API quota).

typescript
1# Add recency filter to reduce redundant queries
2tavily-web-search:
3 days: 1 # only fetch content from last 24 hours for monitoring workflows

Best practices

  • Use Tavily Web Search for automated or scheduled OpenClaw workflows where query volume is high — the lower credit cost per request compared to Tavily AI Search makes sustained automation economically viable on smaller Tavily plans.
  • Pair tavily-web-search with a summarize or analysis skill: collect raw results with Tavily Web Search, then pass the extracted content to a summarization step — this gives you more control over what gets synthesized than the all-in-one Tavily AI Search approach.
  • Set an `exclude_domains` list in your skill config to filter content farms, social aggregators, and low-quality sites from raw results — this is a free quality improvement that does not require AI post-processing.
  • Use the `days` parameter for any monitoring or news-tracking workflow to prevent Tavily from returning old high-ranking content when you only want fresh results.
  • Keep TAVILY_API_KEY stored in ~/.openclaw/.env rather than passed inline — both tavily-web-search and tavily-ai-search read from the same location, and a single secret file is easier to rotate than multiple configuration entries.
  • For bulk URL collection pipelines, set `max_results: 20` (the maximum) and process the full result set downstream rather than making multiple smaller queries — this is more credit-efficient.
  • Monitor your Tavily dashboard usage weekly when running automated workflows to catch quota consumption patterns before you hit the monthly limit and queries start failing.

Alternatives

Frequently asked questions

How do I install Tavily Web Search in OpenClaw?

Run `clawhub install tavily-web-search` in your terminal. Then configure your TAVILY_API_KEY with `clawhub config set TAVILY_API_KEY tvly-your-key`. If you already have tavily-ai-search installed, you can skip the API key step — both skills share the same key.

Do I need a separate API key for Tavily Web Search and Tavily AI Search?

No — both skills use the same TAVILY_API_KEY from your Tavily account at app.tavily.com. You only need one Tavily account and one API key to run both skills simultaneously in OpenClaw.

Why is Tavily Web Search faster than Tavily AI Search?

Tavily Web Search skips the AI post-processing step that Tavily AI Search applies after fetching results. That post-processing — which ranks results by relevance and synthesizes a summary answer — adds 1-3 seconds of latency per query. Tavily Web Search returns the raw extracted content immediately, making it 40-60% faster for the same query.

OpenClaw Tavily Web Search rate limit — what happens when I hit it?

When you exceed the 1,000 monthly free requests, Tavily returns a 429 rate limit error and OpenClaw will show a 'quota exceeded' message. Queries do not queue or retry automatically. You can upgrade your Tavily plan mid-month from the Tavily dashboard to restore quota immediately. Alternatively, configure Brave Search or another search skill as a fallback.

ClawHub install tavily-web-search is not working — what should I check?

Verify the exact skill name is `tavily-web-search` (with a hyphen). Check your internet connection and try again if the registry times out. Run `clawhub install tavily-web-search --force` if a partial install is blocking the fresh installation. If you see a version conflict error, run `clawhub uninstall tavily-web-search` first, then reinstall.

Can RapidDev help set up Tavily Web Search for automated pipelines?

Yes — RapidDev can assist with configuring Tavily Web Search as part of automated OpenClaw monitoring or research workflows, including combining it with other skills for multi-step data collection pipelines. The self-serve setup in this guide covers individual use; teams with complex pipeline requirements can contact RapidDev for configuration support.

RapidDev

Talk to an Expert

Our team has built 600+ apps. Get personalized help with your project.

Book a free consultation

Need help with your project?

Our experts have built 600+ apps and can accelerate your development. Book a free consultation — no strings attached.

Book a free consultation

We put the rapid in RapidDev

Need a dedicated strategic tech and growth partner? Discover what RapidDev can do for your business! Book a call with our team to schedule a free, no-obligation consultation. We'll discuss your project and provide a custom quote at no cost.