<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Francisco García Domene · Prisma Reviews API</title>
    <description>The latest articles on DEV Community by Francisco García Domene · Prisma Reviews API (@fransopa).</description>
    <link>https://dev.to/fransopa</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4211698%2Fa10fc9ec-0d38-430d-93ac-49cfa84f1630.png</url>
      <title>DEV Community: Francisco García Domene · Prisma Reviews API</title>
      <link>https://dev.to/fransopa</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/fransopa"/>
    <language>en</language>
    <item>
      <title>Shipping a remote MCP server with OAuth: what the clients actually check</title>
      <dc:creator>Francisco García Domene · Prisma Reviews API</dc:creator>
      <pubDate>Thu, 12 Nov 2026 11:40:44 +0000</pubDate>
      <link>https://dev.to/fransopa/shipping-a-remote-mcp-server-with-oauth-what-the-clients-actually-check-314n</link>
      <guid>https://dev.to/fransopa/shipping-a-remote-mcp-server-with-oauth-what-the-clients-actually-check-314n</guid>
      <description>&lt;p&gt;Two weeks ago we put a remote MCP server in front of our &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews API&lt;/a&gt;. It lets Claude, Cursor, VS Code and other MCP clients read reviews from the 84 platforms the API covers, and publish a reply, with the user's approval, on the ones whose API allows replies. (How few of them do is the subject of &lt;a href="https://dev.to/fransopa/state-of-review-platform-apis-in-2026-only-8-of-32-let-you-reply-16-have-no-api-at-all-4dnb"&gt;an earlier post&lt;/a&gt;.) The tools were the easy part, because each one forwards to a REST route we already had. Most of the work went into getting clients to connect without a pasted API key.&lt;/p&gt;

&lt;p&gt;These are the things that mattered, in the order we ran into them.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. The 401 is the start of the flow, not an error
&lt;/h2&gt;

&lt;p&gt;A connector UI that supports OAuth doesn't know your login page. It finds it by calling the server with no token and reading the 401 response:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight http"&gt;&lt;code&gt;&lt;span class="no"&gt;HTTP&lt;/span&gt;&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="mf"&gt;1.1&lt;/span&gt; &lt;span class="mi"&gt;401&lt;/span&gt; &lt;span class="nc"&gt;Unauthorized&lt;/span&gt;
&lt;span class="na"&gt;WWW-Authenticate: Bearer resource_metadata&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"https://api.prismareviews.com/.well-known/oauth-protected-resource/mcp"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;scope&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"reviews"&lt;/span&gt;
&lt;span class="nc"&gt;Access&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Control&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Expose&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Headers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="no"&gt;WWW&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Authenticate&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;code&gt;resource_metadata&lt;/code&gt; URL (RFC 9728) points to a small JSON document that names the authorization server. From there the client reads &lt;code&gt;/.well-known/oauth-authorization-server&lt;/code&gt; and gets the authorize, token, registration and revocation endpoints. Ours advertises only &lt;code&gt;authorization_code&lt;/code&gt; with PKCE S256 and public clients (&lt;code&gt;token_endpoint_auth_methods_supported: ["none"]&lt;/code&gt;), because that's all an MCP client needs.&lt;/p&gt;

&lt;p&gt;Two things bit us:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The header has to be exposed to browsers.&lt;/strong&gt; Web clients call the server from a browser, so without &lt;code&gt;Access-Control-Expose-Headers&lt;/code&gt; the client never sees &lt;code&gt;WWW-Authenticate&lt;/code&gt; and gives up.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Don't open the endpoint "to be helpful".&lt;/strong&gt; We were asked to answer &lt;code&gt;initialize&lt;/code&gt; and &lt;code&gt;tools/list&lt;/code&gt; without auth, so that directories could list our tools. We didn't. A 200 on the first call is how a client decides that a server needs no login. If we had returned one, every connector would have skipped OAuth and then failed on the first real call. Instead we publish the tool list, with the input schemas, as a static server card at &lt;code&gt;/.well-known/mcp/server-card.json&lt;/code&gt;. Directories read that; the MCP endpoint itself still answers 401.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  2. Dynamic client registration, without a clients table
&lt;/h2&gt;

&lt;p&gt;MCP clients usually register themselves (RFC 7591) before they start the OAuth flow, and there are many of them. We didn't want a table that grows by one row each time someone opens a new Cursor window. So the flow keeps no state on our side:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The client id is the registration itself.&lt;/strong&gt; It holds the client's name and redirect URIs, signed with HMAC. When the id comes back, we check the signature and read the redirect URIs from it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The authorization code is an encrypted grant.&lt;/strong&gt; It lasts five minutes, needs PKCE S256 and works only once.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The access token is an ordinary API key named after the client.&lt;/strong&gt; It appears in the user's key list as "Cursor (MCP)" or "Claude (MCP)", and revoking it there disconnects that client. The tools keep calling &lt;code&gt;/v1&lt;/code&gt; with a normal key, so the MCP layer doesn't need a second auth path.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Validating redirect URIs took more care than we expected. Desktop and CLI clients come back to &lt;code&gt;http://localhost&lt;/code&gt; or &lt;code&gt;127.0.0.1&lt;/code&gt; on a random port, and editors use their own schemes, such as &lt;code&gt;cursor://&lt;/code&gt; and &lt;code&gt;vscode://&lt;/code&gt;. We allow https, loopback http and custom schemes. We reject plain http on the internet, URLs with a fragment, and &lt;code&gt;javascript:&lt;/code&gt;, &lt;code&gt;data:&lt;/code&gt;, &lt;code&gt;file:&lt;/code&gt; and similar schemes.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Let the user choose test or live on the consent screen
&lt;/h2&gt;

&lt;p&gt;The consent screen asks whether to connect &lt;strong&gt;Live&lt;/strong&gt; (your real accounts) or &lt;strong&gt;Test&lt;/strong&gt; (sample data from every platform). People who are only trying the connector don't need to connect a platform account first. It also gave the directory review a test account with data to look at, instead of an empty one.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Keep a header fallback for clients without OAuth
&lt;/h2&gt;

&lt;p&gt;Claude Code, Cursor and VS Code configs take a header, and some people prefer an API key to OAuth. So the same URL also accepts &lt;code&gt;Authorization: Bearer &amp;lt;key&amp;gt;&lt;/code&gt;. Some gateways keep &lt;code&gt;Authorization&lt;/code&gt; for themselves, so the server also accepts &lt;code&gt;X-API-Key&lt;/code&gt;. Claude Desktop needs &lt;code&gt;mcp-remote&lt;/code&gt; to pass a header to a remote server, so the docs give that config in full.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Stateless JSON over Streamable HTTP
&lt;/h2&gt;

&lt;p&gt;Our tools are request and response: list reviews, get stats, reply. So the server answers each POST with &lt;code&gt;application/json&lt;/code&gt;, with no SSE stream and no session id. That runs well on serverless functions, and no client has needed more. We accept the 2025-06-18, 2025-03-26 and 2024-11-05 protocol versions, because clients upgrade at different speeds.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Tool annotations are part of the contract
&lt;/h2&gt;

&lt;p&gt;We have ten tools. Seven only read, and they're marked &lt;code&gt;readOnlyHint: true&lt;/code&gt;. &lt;code&gt;reply_to_review&lt;/code&gt; is the one that matters:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="n"&gt;annotations&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;readOnlyHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;destructiveHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;idempotentHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;openWorldHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The reply is public and can replace an earlier one, so we mark it &lt;code&gt;destructiveHint&lt;/code&gt;. Clients use that to ask the user before running the tool. &lt;code&gt;openWorldHint&lt;/code&gt; is there because the text leaves our system and appears on the review platform.&lt;/p&gt;

&lt;p&gt;Retries were the other risk. A reader pointed it out in the comments of an earlier post: on platforms where a reply can only be created, never edited, a retry after a dropped connection could publish the same text twice. The REST route under the tool accepts an &lt;code&gt;Idempotency-Key&lt;/code&gt; header, so a repeated request with the same key gets the stored response instead of a second reply.&lt;/p&gt;

&lt;p&gt;We also kept the tool descriptions plain: what the tool does and what it returns. They don't tell the model what to do next. It's easier to keep descriptions honest from the start than to rewrite them for a directory review.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. A tool for the agent that created the account
&lt;/h2&gt;

&lt;p&gt;This one is new since the first version of the server. Some of our tools can't finish a job without the account owner, such as &lt;code&gt;create_connect_link&lt;/code&gt;, which returns a link the owner opens to connect a platform that needs a browser sign-in. But an agent that has created or opened an account for someone can still do the paperwork.&lt;/p&gt;

&lt;p&gt;Every new account answers three sign-up questions: where the owner heard of us, which review platforms they need, and how many locations, apps or listings they have. &lt;code&gt;complete_onboarding&lt;/code&gt; takes those answers over MCP and writes them through &lt;code&gt;PUT /v1/onboarding&lt;/code&gt;, the same route the dashboard form uses:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="n"&gt;annotations&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;readOnlyHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;destructiveHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;idempotentHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;openWorldHint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;platforms&lt;/code&gt; and &lt;code&gt;locations_count&lt;/code&gt; are required.&lt;/strong&gt; &lt;code&gt;origin&lt;/code&gt; is optional. If the agent doesn't know where the owner heard of us, it leaves &lt;code&gt;origin&lt;/code&gt; out and we record the source as an AI agent, or keep the answer the owner already gave. We'd rather store "unknown, came through an agent" than a guess.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It is idempotent.&lt;/strong&gt; Calling it again corrects the answers, so a retry can't make a duplicate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;When the answer is "other", the schema asks for the name.&lt;/strong&gt; &lt;code&gt;origin_detail&lt;/code&gt; and &lt;code&gt;platforms_other&lt;/code&gt; are required in that case, with a length limit.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It's a small tool, and it is the only one that changes the account rather than reading from a platform or publishing to one. That's why it has its own annotations instead of reusing the read-only set.&lt;/p&gt;

&lt;h2&gt;
  
  
  8. Listings
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The official MCP Registry&lt;/strong&gt; verifies a namespace through a key served from your domain. Ours is &lt;code&gt;com.prismareviews/reviews&lt;/code&gt;, and we publish a new &lt;code&gt;server.json&lt;/code&gt; version whenever the tools or the platform list change. It's on 1.0.5 now, and the description names the platform count ("... Viator and 79 more"). The older versions stay in the registry, flagged as not the latest.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude's connector directory&lt;/strong&gt; lists the server. The review wanted a test account with real-looking data, which test mode covered. Any later change to the listing goes back to review, so we finished the listing text before submitting it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Glama&lt;/strong&gt; lets you claim a server through a file at &lt;code&gt;/.well-known/glama.json&lt;/code&gt;, and shows whether the server responds.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What I'd do again
&lt;/h2&gt;

&lt;p&gt;Start with the 401 and the metadata documents, and test them with a real connector UI before writing any tools. If discovery doesn't work, the tools never get called. Of the whole setup, only the redirect URI rules needed real security thought.&lt;/p&gt;

&lt;p&gt;The server is at &lt;code&gt;https://api.prismareviews.com/mcp&lt;/code&gt;. Setup for each client, the tool list and the API route behind each tool are in the docs: &lt;a href="https://prismareviews.com/docs/mcp" rel="noopener noreferrer"&gt;https://prismareviews.com/docs/mcp&lt;/a&gt; (there's also a shorter overview at &lt;a href="https://prismareviews.com/reviews-mcp" rel="noopener noreferrer"&gt;https://prismareviews.com/reviews-mcp&lt;/a&gt;). If you run a remote MCP server and found that clients behave differently from this, I'd like to hear about it in the comments.&lt;/p&gt;

</description>
      <category>mcp</category>
      <category>oauth</category>
      <category>api</category>
      <category>ai</category>
    </item>
    <item>
      <title>Getting a community node verified by n8n in 2026: what it actually took</title>
      <dc:creator>Francisco García Domene · Prisma Reviews API</dc:creator>
      <pubDate>Mon, 02 Nov 2026 15:05:12 +0000</pubDate>
      <link>https://dev.to/fransopa/getting-a-community-node-verified-by-n8n-in-2026-what-it-actually-took-888</link>
      <guid>https://dev.to/fransopa/getting-a-community-node-verified-by-n8n-in-2026-what-it-actually-took-888</guid>
      <description>&lt;p&gt;Built by the team behind &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews API&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;On 2 November 2026 n8n verified our community node for the &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews API&lt;/a&gt;, &lt;a href="https://github.com/Fran-Sopa/n8n-nodes-prisma-reviews" rel="noopener noreferrer"&gt;n8n-nodes-prisma-reviews&lt;/a&gt;. Verified means it shows up in the nodes panel of n8n Cloud and can be dragged onto the canvas with no install step. Here is what the process asked of us, in case you are building one.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Use the official CLI from day one
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;@n8n/node-cli&lt;/code&gt; gives you &lt;code&gt;n8n-node build&lt;/code&gt;, &lt;code&gt;n8n-node lint&lt;/code&gt; and &lt;code&gt;n8n-node release&lt;/code&gt;. The linter is stricter than you expect: options must be sorted alphabetically by name, descriptions follow n8n's style, and credentials need a test request. Fixing lint at the end is painful; run it on every commit (we run it in CI on every push).&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Publish from GitHub Actions with provenance
&lt;/h2&gt;

&lt;p&gt;Since 1 May 2026 n8n only verifies packages published from GitHub Actions with npm provenance. The CLI can scaffold the workflow (&lt;code&gt;n8n-node release --init-workflow&lt;/code&gt;). Our flow now is:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Bump the version and push a tag such as &lt;code&gt;0.1.3&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;The &lt;code&gt;publish.yml&lt;/code&gt; workflow runs lint and build and publishes to npm with a signed provenance statement.&lt;/li&gt;
&lt;li&gt;n8n picks up new versions on its own and reviews them for the next release cycle, as long as the changes are visible in the public repository.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Packages published from a laptop with &lt;code&gt;npm publish&lt;/code&gt; do not get provenance and cannot be verified.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Keep the repository and the package in sync
&lt;/h2&gt;

&lt;p&gt;n8n checks that what is on npm matches the public repo. Every version that goes to npm is a commit and a tag on GitHub first, with a CHANGELOG entry.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Listen to the community thread
&lt;/h2&gt;

&lt;p&gt;The most useful feature in our latest release came from the n8n forum. A user asked how to tell "no new reviews" apart from "the check failed". We added two webhook events to the API, &lt;code&gt;location.sync_failed&lt;/code&gt; (sent once after 3 failed checks in a row) and &lt;code&gt;location.sync_recovered&lt;/code&gt;, and version 0.1.3 of the trigger node exposes them. A workflow can now alert on a broken connection instead of staying silent.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the node does
&lt;/h2&gt;

&lt;p&gt;Prisma Reviews is one API to read and reply to the reviews of businesses and apps you own, on 66 platforms (Google Maps, App Store, Google Play, Judge.me, Yelp, Tripadvisor, Booking.com, G2 and more), in one JSON format. The package has:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Prisma Reviews Trigger&lt;/strong&gt;: starts a workflow on a new or updated review, a published or failed reply, a connection change, or a location whose sync fails or recovers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prisma Reviews&lt;/strong&gt;: an action node to list and get reviews, publish replies where the platform allows it, and manage locations and connections.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Typical first workflows: a Slack message for every new review, an alert to the manager on 1 and 2 star reviews, and a Google Sheet with every review of the month. The node never writes replies itself; it publishes the exact text your workflow passes.&lt;/p&gt;

&lt;p&gt;Setup guide: &lt;a href="https://prismareviews.com/n8n-reviews" rel="noopener noreferrer"&gt;https://prismareviews.com/n8n-reviews&lt;/a&gt; · Node page on n8n: &lt;a href="https://n8n.io/integrations/prisma-reviews/" rel="noopener noreferrer"&gt;https://n8n.io/integrations/prisma-reviews/&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;If you are going through verification and get stuck on lint or provenance, ask in the comments.&lt;/p&gt;

</description>
      <category>n8n</category>
      <category>automation</category>
      <category>opensource</category>
      <category>api</category>
    </item>
    <item>
      <title>State of review platform APIs in 2026: only 8 of 32 let you reply, 16 have no API at all</title>
      <dc:creator>Francisco García Domene · Prisma Reviews API</dc:creator>
      <pubDate>Wed, 28 Oct 2026 19:06:56 +0000</pubDate>
      <link>https://dev.to/fransopa/state-of-review-platform-apis-in-2026-only-8-of-32-let-you-reply-16-have-no-api-at-all-4dnb</link>
      <guid>https://dev.to/fransopa/state-of-review-platform-apis-in-2026-only-8-of-32-let-you-reply-16-have-no-api-at-all-4dnb</guid>
      <description>&lt;p&gt;Built by the team behind &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews API&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;We integrated 32 review platforms into one API. Here is what each one actually lets a developer do, as of October 2026. Full report with the per-platform table: &lt;a href="https://prismareviews.com/guides/state-of-review-platform-apis-2026" rel="noopener noreferrer"&gt;https://prismareviews.com/guides/state-of-review-platform-apis-2026&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The numbers
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Measure&lt;/th&gt;
&lt;th&gt;Count&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Platforms in the report&lt;/td&gt;
&lt;td&gt;32&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Let the owner reply to reviews by API&lt;/td&gt;
&lt;td&gt;8 (25%)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Read only for third parties&lt;/td&gt;
&lt;td&gt;24 (75%)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Expose an official reviews API the owner can authorise&lt;/td&gt;
&lt;td&gt;16 (50%)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;No usable third-party reviews API (public page only)&lt;/td&gt;
&lt;td&gt;16 (50%)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Official API gated behind a paid plan or add-on&lt;/td&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Official API gated behind manual approval of the integrator&lt;/td&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Push notifications for new reviews to third parties&lt;/td&gt;
&lt;td&gt;1 (Google, and only with Cloud Pub/Sub)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Read the first two rows together: if you build a product that "replies to reviews", only a quarter of the platforms your customers care about will let you, and on every one of them you need the owner's own authorisation.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Replies are the exception
&lt;/h2&gt;

&lt;p&gt;The 8 platforms with a reply API: Google Business Profile, App Store, Google Play, Judge.me, Trustpilot, eBay, Trusted Shops and eKomi. Even there the rules differ:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Google caps a reply at 4,096 bytes and 10 edits per minute per profile, and needs the &lt;code&gt;business.manage&lt;/code&gt; OAuth scope.&lt;/li&gt;
&lt;li&gt;Judge.me can create a reply but never edit or delete it by API.&lt;/li&gt;
&lt;li&gt;Apple and Google Play allow create, update and delete, with an App Store Connect key (Customer Support role) or a Play Console service account.&lt;/li&gt;
&lt;li&gt;Trustpilot's reply API comes with its API module, sold to the business, not to the integrator.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The other 24, including Yelp, Tripadvisor, Booking.com, G2 and Capterra, expect the owner to answer in their own dashboard. The honest design for a multi-platform product: a reply endpoint that returns &lt;code&gt;422 reply_not_supported&lt;/code&gt; plus a link to the review on the platform, instead of pretending.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Half the platforms have no reviews API at all
&lt;/h2&gt;

&lt;p&gt;16 of 32 give a third party nothing better than the public listing page:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Never had one:&lt;/strong&gt; OpenTable, DoorDash, Checkatrade, Avvo, Angi, Apartments.com, Capterra, Salesforce AppExchange.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Had one and closed it:&lt;/strong&gt; Facebook removed the Page ratings API in Graph API v22.0; Booking.com paused new connectivity partners; Tripadvisor's Content API returns a handful of reviews for display, not an owner feed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Public read endpoints with no merchant credentials:&lt;/strong&gt; Steam, WordPress.org and Etsy.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Yelp is the edge case: its Fusion API exists, but returns 3 review excerpts per business.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Access is gated by process more than by code
&lt;/h2&gt;

&lt;p&gt;Of the 16 platforms with a usable official API:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Manual approval of the integrator:&lt;/strong&gt; Google Business Profile, Trustpilot, eBay (and Etsy for its public reviews API). Google requires a Business Profile verified for 60+ days before you can even apply.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Paid plan on the business's side:&lt;/strong&gt; Trustpilot (API module), REVIEWS.io (Plus plan), Loox (Convert or Unlimited), ProvenExpert (Premium or Enterprise).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Just credentials:&lt;/strong&gt; App Store Connect, Google Play, Judge.me, Yotpo, G2, Okendo, Feefo, Trusted Shops, eKomi, Shopper Approved.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  4. Nobody pushes
&lt;/h2&gt;

&lt;p&gt;Only Google offers notifications for new reviews, and only through the Notifications API plus a Cloud Pub/Sub topic you run. Everyone else: you poll, within each platform's quota (Google Play: 200 reads per hour per app; Google: 300 QPM shared across the whole project).&lt;/p&gt;

&lt;h2&gt;
  
  
  What this means if you are building on reviews
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Design for read-only first. Replies are a feature of 8 platforms, not of "reviews".&lt;/li&gt;
&lt;li&gt;Budget for approvals in calendar time, not engineering time. Google alone is 60 days of verification plus the review of your application.&lt;/li&gt;
&lt;li&gt;Store history yourself. Google Play returns 7 days; Yelp 3 excerpts; some platforms change their mind (Facebook).&lt;/li&gt;
&lt;li&gt;Normalise early. 32 JSON shapes, 4 rating scales (5 stars, 10 points, thumbs, recommend/not), 3 date formats.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The per-platform table (read, reply, access type, gate, webhooks) is in the full report: &lt;a href="https://prismareviews.com/guides/state-of-review-platform-apis-2026" rel="noopener noreferrer"&gt;https://prismareviews.com/guides/state-of-review-platform-apis-2026&lt;/a&gt;. The access rules for each platform, with the connection body you send, are on &lt;a href="https://prismareviews.com/which-review-platforms-have-an-api" rel="noopener noreferrer"&gt;https://prismareviews.com/which-review-platforms-have-an-api&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If a platform in the list has changed its API since October 2026, tell me in the comments and I will update the report.&lt;/p&gt;

</description>
      <category>api</category>
      <category>webdev</category>
      <category>backend</category>
      <category>saas</category>
    </item>
    <item>
      <title>What I learned integrating 25 review platforms into one API</title>
      <dc:creator>Francisco García Domene · Prisma Reviews API</dc:creator>
      <pubDate>Tue, 27 Oct 2026 20:42:24 +0000</pubDate>
      <link>https://dev.to/fransopa/what-i-learned-integrating-25-review-platforms-into-one-api-4068</link>
      <guid>https://dev.to/fransopa/what-i-learned-integrating-25-review-platforms-into-one-api-4068</guid>
      <description>&lt;p&gt;Built by the team behind &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews API&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Every review platform looks the same from the outside: a star rating, a text, an author, maybe a reply. Getting at that data looks nothing alike. I spent the last months building &lt;a href="https://prismareviews.com" rel="noopener noreferrer"&gt;Prisma Reviews&lt;/a&gt;, a single REST API for reviews, and these are the notes I wish I'd had on day one.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. There are four kinds of "review API"
&lt;/h2&gt;

&lt;p&gt;Once you've done a few platforms, every one falls into one of four kinds:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Kind&lt;/th&gt;
&lt;th&gt;Examples&lt;/th&gt;
&lt;th&gt;How you connect&lt;/th&gt;
&lt;th&gt;Can you reply?&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;OAuth, business-owner API&lt;/td&gt;
&lt;td&gt;Google Business Profile, Trustpilot&lt;/td&gt;
&lt;td&gt;The owner approves your app&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Keys from a developer console&lt;/td&gt;
&lt;td&gt;App Store Connect, Google Play&lt;/td&gt;
&lt;td&gt;The owner creates a key and gives it to you&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Shop-app tokens&lt;/td&gt;
&lt;td&gt;Judge.me, Trusted Shops, eKomi, Okendo&lt;/td&gt;
&lt;td&gt;An API token from the platform's admin&lt;/td&gt;
&lt;td&gt;Sometimes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;No owner API at all&lt;/td&gt;
&lt;td&gt;Yelp, Tripadvisor, Booking.com&lt;/td&gt;
&lt;td&gt;Only the public listing URL&lt;/td&gt;
&lt;td&gt;No: link to the platform&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Each kind needs its own onboarding UI, its own error handling and its own way to refresh credentials. If you design for only one, the next one won't fit.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Access is the real bottleneck, not code
&lt;/h2&gt;

&lt;p&gt;The code for a connector is usually a day or two. Getting permission takes far longer:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Google Business Profile&lt;/strong&gt; needs an access application. It's reviewed by hand, and the default quota is very low until you're approved.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Trustpilot&lt;/strong&gt; sells API access as a paid add-on to the business, not to you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;eBay&lt;/strong&gt; and &lt;strong&gt;Etsy&lt;/strong&gt; need a developer app per integrator, with their own review of your use case.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Apple&lt;/strong&gt; and &lt;strong&gt;Google Play&lt;/strong&gt; are the easy ones: the owner creates a key in their console (an App Store Connect &lt;code&gt;.p8&lt;/code&gt; key, a Play service account) and you can start right away.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The lesson: ship the platforms with self-serve keys first, apply for the gated ones on day one, and be honest in your UI about which ones are still waiting.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Normalise early, and keep the original
&lt;/h2&gt;

&lt;p&gt;Every platform has its own field names (&lt;code&gt;starRating: "FOUR"&lt;/code&gt;, &lt;code&gt;rating: 4&lt;/code&gt;, &lt;code&gt;stars: "4.0"&lt;/code&gt;), dates (ISO, Unix seconds, &lt;code&gt;"2026-10-19 12:41:07 +00:00"&lt;/code&gt;) and ideas of what an "author" is. We map everything into one shape:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"id"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"rev_01KJ8Q2M4T7HX3VB9CWD6NRZ5E"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"app_store"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"rating"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;4&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Great, one bug"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Love the app, the widget crashes on iOS 19."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"author"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"marta_g"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"language"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"en"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"created_at"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"2026-10-19T12:41:07.000Z"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"reply"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Thanks! Fixed in 3.2."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"status"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"published"&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Two rules that saved us later:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Ratings are integers 1–5 or &lt;code&gt;null&lt;/code&gt;.&lt;/strong&gt; Some platforms allow reviews with no rating, so a missing rating is &lt;code&gt;null&lt;/code&gt;, never &lt;code&gt;0&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prefer the original text over the translation.&lt;/strong&gt; Several sources auto-translate reviews to the reader's language. Store what the customer actually wrote and its language.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  4. Polling is fine if you're honest about cost
&lt;/h2&gt;

&lt;p&gt;Most platforms have no webhooks, so you poll. That's fine for official APIs, where calls are free and rate limits are generous. It's not fine when every read costs you money (scraped sources charge per review). What worked for us:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Poll official APIs often (every 15–60 minutes).&lt;/li&gt;
&lt;li&gt;Poll paid sources rarely (every 6–24 hours), always ask for "newest first", and stop at the first review you've already seen.&lt;/li&gt;
&lt;li&gt;Reserve a budget before every paid job and cap the job at that budget, so a bug can't spend a month's margin in an hour.&lt;/li&gt;
&lt;li&gt;Expose one webhook (&lt;code&gt;review.created&lt;/code&gt;, &lt;code&gt;review.updated&lt;/code&gt;) to your users, whatever the platform does underneath.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  5. Replies are a queue, not a request
&lt;/h2&gt;

&lt;p&gt;Publishing a reply can take seconds (Apple), fail because the review was deleted, or be rate-limited. So &lt;code&gt;POST /reviews/{id}/reply&lt;/code&gt; returns at once with &lt;code&gt;status: "queued"&lt;/code&gt;, a worker publishes it, and a &lt;code&gt;reply.published&lt;/code&gt; or &lt;code&gt;reply.failed&lt;/code&gt; webhook tells you how it went. Idempotency keys stop a retry from posting the same reply twice, and a customer reading the review a second later sees a queued reply instead of none.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Make it testable without a real account
&lt;/h2&gt;

&lt;p&gt;The biggest friction for a developer trying a reviews API is "I need a real business with real reviews first". Test keys (&lt;code&gt;pk_test_…&lt;/code&gt;) that return realistic sample data for every platform let someone build the whole integration before connecting anything real. They're also what we use in our own test suite.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Agents are users too
&lt;/h2&gt;

&lt;p&gt;A surprising share of our early traffic was AI agents reading docs. Three cheap things helped:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Every docs page is also available as Markdown (add &lt;code&gt;.md&lt;/code&gt; to the URL), and there's an &lt;code&gt;llms.txt&lt;/code&gt; file.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;GET /v1/platforms&lt;/code&gt; returns, for each platform, which fields to send, the steps the owner follows and an example body. An agent can connect a platform without reading the docs.&lt;/li&gt;
&lt;li&gt;Error responses include &lt;code&gt;required_fields&lt;/code&gt; and an &lt;code&gt;example&lt;/code&gt;, so the next request can fix itself.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Wrapping up
&lt;/h2&gt;

&lt;p&gt;If you're integrating reviews from more than two platforms, the time goes into access, normalisation and cost control, not HTTP calls. That's the work we packaged into &lt;a href="https://prismareviews.com/docs" rel="noopener noreferrer"&gt;Prisma Reviews&lt;/a&gt;. It's free for 2 locations, with test keys for every platform. If you'd rather build it yourself, I hope the notes above save you a few weeks.&lt;/p&gt;

&lt;p&gt;Questions about a specific platform's API? Ask in the comments. I've probably hit the same wall.&lt;/p&gt;

</description>
      <category>api</category>
      <category>webdev</category>
      <category>javascript</category>
      <category>saas</category>
    </item>
  </channel>
</rss>
