SerpApi yesterday made its search-data tool downloadable as a single file that installs itself into Claude Desktop with a drag and drop. Once installed, anyone chatting with Claude can ask it to pull live Google, Bing, Amazon or YouTube results into the conversation, instead of Claude relying only on what it already knows. The practical change is that setup, which used to mean editing configuration files by hand, now takes a few clicks and one pasted API key.

SerpApi, the Austin, Texas-based search-data provider, yesterday announced support for the MCP Bundle packaging format in its Model Context Protocol server, allowing the tool to be installed in Claude Desktop and other compatible applications as a single downloadable file rather than through manual configuration. According to a post by Python Developer Advocate Adarsh Divakaran published on the SerpApi blog, the bundle can be downloaded from the company's GitHub Releases page and dropped directly onto the Extensions settings screen in Claude Desktop, where the application handles installation and prompts the user for an API key.

The release extends a product that already sits inside one of the most closely watched disputes in search infrastructure. SerpApi is currently defending itself against a DMCA lawsuit brought by Google and a separate suit from Reddit, both alleging that the company circumvents anti-bot measures to harvest search results at scale. Thursday's announcement does not touch that litigation. It addresses a narrower, largely technical question: how a locally running search tool gets onto a user's machine and into an AI assistant's toolset.

What MCP Bundles are

The Model Context Protocol, the open standard Anthropic introduced in November 2024 for connecting AI assistants to external tools and data sources, defines how a language model discovers and calls a server's capabilities. It does not, on its own, define how that server gets installed on a user's computer. MCP Bundles, abbreviated MCPB, fill that gap. According to the SerpApi post, each .mcpb file is a ZIP archive containing a manifest.json file at its root, and that manifest describes the server, what it can do, and how the host application should launch it.

The manifest specification also covers configuration fields and compatibility requirements, the post explains. A server can declare that it requires an API key, for instance, so the host application knows to collect one from the user during setup. Developers can specify which operating systems and runtime versions their server supports, allowing an installer to reject an incompatible bundle before it causes a failure later.

MCPB is described in the post as an open packaging format that other applications can implement, with the official repository holding the specification, packaging tools and the code Claude Desktop itself uses to load bundles. The post is explicit that support for MCP as a protocol does not automatically translate into support for the bundle format: an application also needs to understand the specific manifest version and server runtime a given bundle uses. That distinction separates two layers that are easy to conflate. One is whether an AI application can talk to external tools at all. The other is whether it can install one of those tools without a developer's help.

SerpApi's own bundle runs the MCP server locally on the user's machine and sends search requests to SerpApi using the user's account, according to the announcement. It packages the search tools and engine parameter schemas the server needs, along with the configuration required to start it, and the company says the bundle manifest can be inspected directly in its repository.

Installing SerpApi in Claude Desktop

The post walks through a four-step process for adding the SerpApi bundle to Claude Desktop, framed as requiring the latest version of the application and a SerpApi account with an API key. First, the user downloads the bundle from the SerpApi MCP GitHub Releases page, selecting the asset ending in .mcpb, such as the version the post names as an example, serpapi-mcp-1.0.3.mcpb. Second, the user opens Claude Desktop and navigates to Settings, then Extensions.

Third, the downloaded file is dragged onto the Extensions page, which triggers the installation prompts. When Claude requests the SerpApi API key, the user pastes it from the SerpApi dashboard into the field provided. According to the post, Claude stores that key as a sensitive setting and supplies it to the server as an environment variable named SERPAPI_API_KEY, rather than embedding it in a configuration file the user would otherwise have to edit and protect manually. The post notes an alternative installation route through Advanced Settings and an Install Extension option, and adds that users should confirm the SerpApi extension is enabled from the extension settings after installation completes.

Fourth, the user starts a new conversation and asks Claude to use SerpApi, with the post giving example prompts such as searching Google for coffee shops in Austin or looking up academic papers through Google Scholar. Claude may ask for permission before running the SerpApi tool, and once granted, follow-up questions about the results can continue in the same conversation. The post adds that the bundle also includes SerpApi's separate MCP Apps tools, which render interactive search tables and dashboards in compatible hosts, a capability the company says it previously covered in a post titled "MCP Apps with FastMCP: Turning Tool Output Into Interactive UI."

Building a bundle from scratch

A separate portion of the announcement is addressed to developers who maintain their own MCP servers rather than to marketers evaluating SerpApi as a research tool. For an existing local server, the post recommends starting with the official MCPB command-line interface, describing an interactive command called mcpb init that collects a server's metadata, entry point and configuration before dependencies are prepared for whichever runtime is chosen.

According to the packaging guidance the post cites, Node.js is recommended for reducing installation work specifically in Claude Desktop, because that application already includes a Node.js runtime. The format also supports Python and compiled executables as server types, each carrying different packaging considerations. For Node.js servers, the guidance recommends reproducible dependency installation using commands such as npm ci. Python bundles built the traditional way need care with compiled dependencies, since those can vary by operating system and processor architecture. For compiled binaries, the recommendation favors static linking or bundling required shared libraries, followed by testing on a clean system that has none of the developer's local tooling installed.

A .mcpbignore file can exclude development files from the final archive, mirroring the function of a .gitignore file in version control. Once the runtime files are ready, the manifest is validated and the archive created using two further CLI commands: mcpb validate manifest.json, then mcpb pack, pointing at the source directory and naming the output file. The post recommends installing the finished bundle in the target application afterward and confirming that it starts, accepts its configuration and runs at least one tool successfully, distinguishing between manifest validation, which checks structure, and an installation check, which exercises the packaged server directly.

How SerpApi packages its own server

SerpApi's server is written in Python and uses the uv runtime, according to the post. The bundle includes the source code, engine schemas, a pyproject.toml file and a uv.lock file. Its launch command runs as uv run with the frozen and no-dev flags, meaning the host installs the exact locked dependency versions without pulling in development-only dependencies, and the company says it keeps installed packages and local virtual environments out of the archive entirely.

The API key field in the manifest is declared as both required and sensitive, and its value reaches the server through the SERPAPI_API_KEY environment variable, which the company's local entry point reads when making search requests. SerpApi's build script prepares the archive, checks its contents and starts the packaged server over stdio for what the post calls a smoke check, a basic functional test confirming the bundle runs before it ships. The release workflow then attaches the resulting .mcpb file to a GitHub release automatically. The post points developers toward build instructions in the repository's MCP Bundle documentation for further detail, and invites feedback, feature suggestions or bug reports through the SerpApi MCP repository.

Context: a company mid-litigation

The bundle-support announcement lands against the backdrop of an active legal fight that has shaped how SerpApi is covered and understood over the past year. Google filed a DMCA lawsuit against SerpApi on December 19, 2025, alleging that the company circumvents SearchGuard, the JavaScript-based bot-detection system Google deployed in January 2025, to scrape search results on what the original complaint described as billions of separate occasions. SerpApi filed a motion to dismiss on February 20, 2026, arguing that the DMCA is a copyright statute rather than a mechanism for controlling access to publicly available information, and that Google does not hold a copyright in search results it merely displays.

That motion succeeded in large part. A federal judge dismissed every claim in Google's original complaint on July 20, 2026, ruling that SearchGuard does not protect copyrighted work in the way the DMCA requires, though the court left Google a narrower path to amend its case around specific licensed content such as photographs inside Knowledge Panels. Google took that path, and PPC Land's coverage of the amended filing noted that the case now rests entirely on whether Google can show its content licensors authorized SearchGuard to guard their material, a considerably narrower theory than the one the case began with. Separately, Reddit sued SerpApi and three other companies in October 2025 over similar circumvention allegations tied to Reddit content appearing in Google search results, and SerpApi pushed to have that case dismissed in March 2026 as well.

None of that litigation concerns the MCP Bundle release directly, and the announcement itself makes no mention of the lawsuits. But the timing places the product update inside a period when SerpApi has also had to respond quickly to changes on Google's side of the relationship. Earlier this month, SerpApi restored direct destination URLs after Google rolled out google.com/goto redirect links that had concealed where a search result actually pointed until a user clicked it, a change the company said required updates to how result links were decoded across multiple result types, since a single parsing method could not resolve every variation.

Why this matters for search-data workflows

MCP adoption across advertising and marketing technology has moved quickly since Anthropic published the protocol. Google released an open-source MCP server for its Ads API in October 2025. Amazon Ads opened a closed beta of its own server the following month and moved to open beta in February 2026. Meta opened its ads system to Claude and ChatGPT through MCP connectors with write access in April 2026, and by July 2026 had removed custom integration requirements for third-party developers entirely. Snap and Pinterest followed with their own MCP servers in the same window. The pattern across nearly all of these launches has been the same: a platform exposes a server, and a developer or marketer configures a client to talk to it, typically by editing a JSON configuration file by hand.

SerpApi's bundle-support release does not change what its server does. It changes how a non-developer gets it running. That distinction matters for the population of marketers, SEO practitioners and competitive-intelligence analysts who use SerpApi's data feeds but do not write code for a living. A rank tracker or a person auditing search-results layouts across markets can, under the process described in the announcement, install live Google, Bing, Amazon and YouTube search access inside a Claude conversation without touching a configuration file, provided they have Claude Desktop and a SerpApi API key. Anthropic's own Connectors Directory, which the announcement references as one destination for MCPB packaging, represents a distribution channel that a handful of the largest ad platforms have not yet used for their own MCP servers, most of which still require manual client-side setup.

The announcement also illustrates a coordination problem that the wider MCP ecosystem has not fully resolved. As the post states plainly, an application's ability to connect to MCP servers in general does not mean it can install .mcpb files specifically, since that additionally requires support for a given bundle's manifest version and server runtime. Claude Desktop supports the format today. Other MCP-compatible applications, including code editors and other AI assistants that have adopted the protocol, may or may not support MCPB installation, leaving developers to decide whether packaging effort spent on one host's installer format is worth duplicating for others. That fragmentation risk sits underneath an otherwise straightforward usability improvement, and it is one reason the post frames MCPB as an open format other applications "can" implement rather than one they already have.

For a publication that tracks how programmatic infrastructure and AI-driven tools change the daily mechanics of digital marketing work, the SerpApi release is a small but concrete data point in a larger trend: AI assistants are absorbing search-data tooling that previously required a separate dashboard, browser tab, or API client, and the barrier to that absorption is increasingly packaging and distribution rather than the underlying protocol.

Timeline

Summary

Who: SerpApi, the Austin, Texas-based search-data API provider, through Python Developer Advocate Adarsh Divakaran, addressing developers and marketers who use Claude Desktop and other MCP-compatible applications.

What: SerpApi added MCP Bundle (MCPB) packaging support to its Model Context Protocol server, allowing the server to be installed as a single .mcpb file through Claude Desktop's Extensions settings rather than through manual configuration, alongside published guidance for developers packaging their own MCP servers in the same format.

When: The announcement was published today, September 9, 2026 (per the source document's byline), with the bundle available for download from SerpApi's GitHub Releases page.

Where: The release applies to Claude Desktop as the primary supported application, with the MCPB format itself designed to be implementable by any compatible desktop AI application; SerpApi is based in Austin, Texas.

Why: The change lowers the technical barrier for installing search-data tools inside AI assistants, replacing manual JSON configuration with a drag-and-drop installer and an API-key prompt, at a time when MCP adoption across search and advertising platforms has been rapid but has generally still required developer-level setup.