Yelp MCP Server

A hosted Model Context Protocol (MCP) server that gives Claude, Cursor, Windsurf and any other MCP client three read-only Yelp tools. Search businesses by keyword and location, read one business in full, and page through its complete review feed, all as structured JSON, with no Yelp Fusion key and nothing to host.

It reads public Yelp pages that a signed-out visitor can see, on any of the 41 regional domains.

1,000 free credits every month, no card required, which is 100 Yelp calls at the 10-credit rate.

https://mcp.hasdata.com/api/mcp?apis=yelp

Contents

• What you need

• Quick start

• Example prompts

• Tools

• Errors and failure paths

• Pricing, free tier and limits

• Tool selection

• How it compares

• FAQ

• HasData links

• Development

• Contributing

• License

What you need

An MCP client and a HasData API key from the dashboard, free to create with no card, and the free tier covers about 100 calls a month at the 10-credit rate. This is a remote server, so the simplest path is a URL and an x-api-key header, with no container to run. A client that only speaks stdio reaches it through a thin launcher, published as @hasdata/yelp-mcp on npm and hasdata-yelp-mcp on PyPI, shown below.

Quick start

The server URL is the same for every client. We run it hands-on in Claude Code and Claude Desktop. The other blocks follow each client's own documented format for a remote server.

Clients with OAuth support can add the same URL as a connector and sign in without putting a key in a config file.

claude mcp add --transport http yelp "https://mcp.hasdata.com/api/mcp?apis=yelp" \
  --header "x-api-key: HASDATA_API_KEY"

Settings, then Connectors, then Add custom connector, then paste https://mcp.hasdata.com/api/mcp?apis=yelp and sign in.

For the config-file route, Claude Desktop loads only local (stdio) servers, so it reaches a remote server through a stdio launcher. The @hasdata/yelp-mcp package is that launcher, and it reads the key from the environment. Add this to claude_desktop_config.json:

{
  "mcpServers": {
    "yelp": {
      "command": "npx",
      "args": ["-y", "@hasdata/yelp-mcp"],
      "env": { "HASDATA_API_KEY": "YOUR_KEY" }
    }
  }
}

For Python instead of Node, swap the launcher for the PyPI package, which uvx runs without a manual install:

{
  "mcpServers": {
    "yelp": {
      "command": "uvx",
      "args": ["hasdata-yelp-mcp"],
      "env": { "HASDATA_API_KEY": "YOUR_KEY" }
    }
  }
}

~/.cursor/mcp.json for every project, or .cursor/mcp.json for one:

{
  "mcpServers": {
    "yelp": {
      "url": "https://mcp.hasdata.com/api/mcp?apis=yelp",
      "headers": { "x-api-key": "HASDATA_API_KEY" }
    }
  }
}

~/.codeium/windsurf/mcp_config.json. Windsurf calls the field serverUrl, not url:

{
  "mcpServers": {
    "yelp": {
      "serverUrl": "https://mcp.hasdata.com/api/mcp?apis=yelp",
      "headers": { "x-api-key": "HASDATA_API_KEY" }
    }
  }
}

.vscode/mcp.json in the workspace:

{
  "servers": {
    "yelp": {
      "type": "http",
      "url": "https://mcp.hasdata.com/api/mcp?apis=yelp",
      "headers": { "x-api-key": "HASDATA_API_KEY" }
    }
  }
}

Example prompts

Each of these lands on one tool, or on two in sequence when the second needs an identifier the first returns.

• Find coffee roasteries in Austin, TX and rank them by rating against review count.

• Pull the hours, phone and website for the Yelp business desnudo-coffee-austin.

• Read every one-star and two-star review of this business and group the complaints by theme.

• Which dishes does Yelp list as popular for this place, and what do reviewers say about them?

• Compare the rating distribution of these three competitors in the same neighborhood.

• Show me the reviews Yelp does not recommend for this business and how they differ from the recommended ones.

A prompt that names a business rather than a Yelp ID takes two calls, one search to resolve the ID and one place or reviews lookup to read it. The search tool returns both placeId and placeAlias, and either one works as the placeId argument to the place tool.

Tools

Three tools, 10 credits per successful call. Every tool accepts domain to switch country, one of 41 values, from www.yelp.com through the European, Asian and Latin American sites, including the language-specific variants such as fr.yelp.ca and zh.yelp.com.hk.

Get Yelp search results

hasdata_yelp_search_getSearchResults

A page of businesses for a keyword in a place.

Returns searchInformation with the echoed keyword, location and totalResults, an ads array of paid placements, an organicResults array, and pagination with currentPage, perPage, totalPages, nextPageUrl and otherPagesUrls.

The two result arrays are not the same shape. An organic result carries position and streetAddress, while an ad carries neither and adds phone and a highlights array of the badges Yelp shows advertisers. Merging the arrays without checking which one a business came from turns paid placement into rank.

{
  "position": 2,
  "placeId": "oiJ7QuhhpsEpe9zFTZF0bA",
  "placeAlias": "desnudo-coffee-austin",
  "url": "https://www.yelp.com/biz/desnudo-coffee-austin",
  "title": "Desnudo Coffee",
  "streetAddress": "2505 Webberville Rd, Austin",
  "price": "$$",
  "categories": [{ "title": "Coffee Roasteries", "url": "https://www.yelp.com/search?find_desc=Coffee+Roasteries&find_loc=Austin%2C+TX" }],
  "snippet": "My favorite coffee shop to go to ever! Everyone must go. [[HIGHLIGHT]]Great coffee[[ENDHIGHLIGHT]], great vibes,great customer...",
  "rating": 4.7,
  "reviews": 347,
  "thumbnail": "https://s3-media0.fl.yelpcdn.com/bphoto/0nE_IcbRiyxrw3kk2z6owg/ls.jpg",
  "allImagesUrl": "https://www.yelp.com/biz_photos/oiJ7QuhhpsEpe9zFTZF0bA"
}

Get Yelp place details

hasdata_yelp_place_getPlaceDetails

One business in full.

Returns a placeResult object with name, url, address, neighborhoods, country, phone, website, price, categories, rating, reviews, isClaimed and isClaimable, an operationHours object holding a week of hours plus today, a features array, a menu object with popularDishes, a faqs array of questions Yelp readers asked and answered, a reviewHighlights array of the phrases Yelp pins to the top of the page, an images array, and businessMap, a static map image URL.

features is the amenities block, and each entry carries a title and an isActive flag, so a false entry is Yelp stating the amenity is absent rather than unknown. That difference matters when you filter, because dropping the false entries and dropping the missing ones are not the same query.

{
  "name": "Desnudo Coffee",
  "url": "https://www.yelp.com/biz/desnudo-coffee-austin",
  "address": "2505 Webberville Rd Austin, TX 78702",
  "neighborhoods": "East Austin",
  "country": "US",
  "phone": "(424) 400-1857",
  "website": "http://www.desnudocoffee.com",
  "price": "$$",
  "categories": ["Coffee Roasteries"],
  "rating": 4.7,
  "reviews": 347,
  "isClaimed": true,
  "isClaimable": false,
  "operationHours": { "hours": [{ "day": "Mon", "hours": ["7:00 AM - 2:00 PM"] }] },
  "features": [
    { "title": "Offers delivery", "isActive": true },
    { "title": "ADA-compliant restroom", "isActive": false }
  ],
  "menu": {
    "section": "Popular Drinks",
    "popularDishes": [{ "name": "Brown Sugar Miso Latte", "rating": 4.7, "reviews": 113, "photos": 71 }]
  }
}

Get Yelp place reviews

hasdata_yelp_reviews_getPlaceReviews

The review feed of one business, in full text, with sorting, filtering and pagination.

Returns searchInformation with the business name, alias, URL, totalResults, rating, a reviewCountsByRating array and a reviewCountsByLanguage breakdown, a pagination object, and a reviews array.

Each review carries position, id, link, a user object, a comment object holding text and its detected language, date, rating, and, when the reviewer attached them, photos, videos and reactions. The user object reports name, userId, address, lifetime reviews, friends and photos counts, and eliteYear for a Yelp Elite member.

A review the author later rewrote also carries previousReviews, holding the earlier version with its own text, rating and date. That is the field to read when the question is whether a rating moved, because the current review alone cannot answer it.

{
  "position": 1,
  "id": "7zLIm3c2v2hRBVmgaKkR7w",
  "link": "https://www.yelp.com/biz/desnudo-coffee-austin?hrid=7zLIm3c2v2hRBVmgaKkR7w",
  "user": {
    "name": "Karson S.",
    "userId": "3LxSs_dQ37-LBRz07EDbyg",
    "address": "Austin, TX",
    "reviews": 374,
    "friends": 61,
    "photos": 875,
    "eliteYear": "26"
  },
  "comment": { "text": "Coming back to Desnudo to update my old review...", "language": "en" },
  "date": "2026-08-20T17:33:47-05:00",
  "rating": 5,
  "photos": [{ "link": "https://s3-media0.fl.yelpcdn.com/bphoto/YbjFKmRf7j0ejQQSaqtqVw/o.jpg", "caption": "Matcha latte", "width": 1126, "height": 2000 }],
  "reactions": [{ "type": "HELPFUL", "label": "Helpful", "count": 1 }],
  "previousReviews": [{ "id": "0WNI2IG7K1_Dg9zdXm03SA", "rating": 4, "comment": { "text": "..." } }]
}

Errors and failure paths

Plan for these rather than assuming a happy path.

A search with no matches returns a successful result with an empty organicResults array, not an error. requestMetadata.status is still ok. Test the array length before iterating.

snippet is marked-up text, not clean text. Yelp wraps the matched words in [[HIGHLIGHT]] and [[ENDHIGHLIGHT]], and those markers arrive verbatim. Strip them before you index, embed or display the snippet.

Category URLs arrive HTML-escaped. The url inside a categories entry contains & rather than a bare ampersand, because that is how it sits in the page. Unescape it before following the link.

rating and query do not combine on the reviews tool. Yelp ignores the star filter while a free-text query is running, so a filtered search comes back with reviews of every rating. Filter the result yourself when you need both.

start and nextPageToken are two different ways to page, and they do not mix. Pass one or the other. The token carries both the offset and the page size, so resending it alone continues the feed, while start needs num to stay put across calls.

The not-recommended feed is a different feed with different limits. Setting notRecommended returns reviews Yelp filtered out of the main list, and those come ten at a time rather than 49, with no photos, videos or reactions attached.

A business can be unclaimed, and an unclaimed page is thin. isClaimed false usually means no website, no hours and no amenities, because nobody filled them in. Read the flag before you treat a missing field as a scraping failure.

Results that carry data also carry a requestMetadata.id worth quoting in support.

Pricing, free tier and limits

Each Yelp tool costs 10 credits per successful call. Response size does not change the price, so a 49-review page and a 5-review page cost the same, which makes the largest page the cheapest way to read a feed.

The free tier is 1,000 credits every month with no card, which is 100 Yelp calls at the base rate. It renews with the billing cycle, so a low-volume agent runs on the free tier indefinitely.

Paid plans start at $49 a month for 200,000 credits, which is 20,000 calls. The unit price falls with volume, from $2.45 per 1,000 calls on the entry plan to $1.00 on Business, $0.84 on Growth and $0.74 on the largest high-volume plans.

Your plan also sets concurrency. The free tier allows 1 request at a time, Startup 15, Business 30, Growth 50, and the high-volume plans run from 200 to 1,500. Handle the overflow case defensively in anything unattended, because an agent that fans out across a list of businesses will reach the ceiling before you do.

A request that comes back non-200 is not billed. A successful call that finds nothing is still a call.

Tool selection

Start from what the prompt gives you. A keyword and a place go to the search tool, a Yelp ID or alias goes straight to the place tool, and a question about what customers said goes to the reviews tool. Spending a search call to reach an ID you already have is the most common waste.

Then pick by what the question is about. The place tool answers questions about the business itself, its hours, amenities, price tier and headline rating. The reviews tool answers questions about its customers, and it is the only one that returns review text, authors and the rating distribution. The reviewHighlights block on the place tool is a sample Yelp curates, not a substitute for the feed.

Read the feed with the largest page. num defaults to its maximum of 49 already, so leave it alone unless you are deliberately sampling.

How it compares

Yelp's own Fusion API is the official route to this data, and it is a different instrument.

The row that decides it is review text. Fusion returns three excerpts per business, which answers a display question on a storefront and cannot answer an analysis question about sentiment, complaints or how a rating moved. When three excerpts and an official contract are what you need, Fusion is the better fit.

FAQ

Is there an official Yelp MCP server?

Yelp does not publish one. This one is maintained by HasData and reads public Yelp pages.

What is a Yelp MCP server?

An MCP server exposes tools an AI client can call. This one turns Yelp search results, business pages and review feeds into JSON an agent can reason over, without a browser or a scraping library in your stack.

Do I need a Yelp account or a Fusion key?

No. The only credential is your HasData key.

Which Yelp sites are covered?

All 41 domains the API accepts, from www.yelp.com through the European, Asian and Latin American sites. Several countries have more than one, split by language, such as fr.yelp.ca next to www.yelp.ca. Pass domain to switch.

Can I get every review of a business?

Yes, by paging. The feed returns 49 at a time, and pagination.hasNextPage tells you when to stop. Reading a business with 347 reviews takes eight calls.

What is the difference between recommended and not-recommended reviews?

Yelp runs software that hides some reviews from the main feed. The default response is the recommended feed, the one a visitor sees. Setting notRecommended returns the hidden one instead, which is smaller, paged ten at a time, and stripped of photos and reactions.

Why did my rating filter return every rating?

Because a query was set at the same time. Yelp drops the star filter when it runs a text search, so the two cannot be combined server-side.

Can I use this together with other HasData APIs?

Yes. One key covers everything, and one endpoint serves them all through the apis parameter. Point a client at ?apis=yelp,google_maps to get both tool sets in one connection, or at mcp.hasdata.com/api/mcp for the full catalogue.

Is HasData affiliated with Yelp?

No. HasData is an independent service and is not affiliated with, endorsed by, or sponsored by Yelp. Yelp is a trademark of its respective owner. The tools work with publicly available data only, and you are responsible for using the results in line with Yelp's terms and the law that applies to you.

Compliance and personal data

The review tools return personal data. A review carries the author's display name, profile photo, stated location, user ID and a link to their profile, and reviewers are private individuals rather than businesses. That puts the response in scope of the GDPR and the CCPA in a way a business listing is not. Decide what you actually need before you store it, keep it no longer than the purpose requires, and check your own obligations. Aggregate analysis rarely needs the author fields at all.

HasData links

• Yelp Scraper API, the REST endpoints behind these tools

• API documentation

• MCP server documentation

• Pricing

• Dashboard

Other HasData MCP servers: Google Search, Google Maps, Google Trends, Google Flights, DuckDuckGo, YouTube, TikTok, Instagram, Amazon, Zillow, Airbnb, Booking.com, Indeed.

Development

The launcher is a thin stdio bridge to the remote server, so there is nothing to build.

npm install
HASDATA_API_KEY=your_key_here npm test

The tests in test/ assert the tool contract, the part that can break without a commit here. They check that ?apis=yelp returns the expected tool count, that no name changed, that every tool still declares its required parameters and carries a description, and that the key in use is actually accepted. That last check calls a tool for real and costs 10 credits, which is the price of a canary that can fail for the right reason.

The contract suite also runs weekly on a schedule, because the upstream tool list can change without anyone touching this repository.

Contributing

A tool table, a response sample or a documented behaviour that does not match reality is worth an issue. There is a template for exactly that. Pull requests are welcome for the same, and for anything in the launcher.

License

MIT, see LICENSE.