Python MCP Packages Out-Downloaded Ten AI Agent Frameworks Combined in Q3 2026 and Reached 111,279,292 Downloads in a Single Week, About 20 Months After MCP Was Introduced

Summary

Picture a dental office whose scheduling software now offers an "AI connector." Behind that button is often the Model Context Protocol, or MCP, a standard way for AI assistants to plug into outside tools and data, from calendars to inventory systems to websites.

Developers build MCP servers and clients in Python with two main packages, the official mcp SDK and fastmcp. Both are public on the Python Package Index, so their download counts give a rough view of how quickly this plumbing for AI agents is spreading.

Very quickly. MCP was introduced in late November 2024. In the first week of August 2026 the two packages were downloaded 111,279,292 times. Across the Q3 2026 sample days they passed the combined downloads of ten popular agent frameworks.

What we measured

We queried the public PyPI download log in Google BigQuery (bigquery-public-data.pypi.file_downloads), which records one row for every file downloaded from the Python Package Index. To keep the query affordable we sampled the first seven days of every month from January 2022 through September 2026, and summed downloads by month, package, country, CI flag, installer and weekday versus weekend into one working table. A quarter in this study therefore means 21 sampled days (days 1 to 7 of each of its three months), not a full quarter. We removed downloads made by mirroring and proxy tools (bandersnatch, Nexus, Artifactory, devpi), which copy packages in bulk rather than install them. To control for the growth of PyPI itself we also pulled total daily download counts for the whole index from the table's partition metadata for the same sampled days.

We tracked two packages: mcp, the official Python SDK, and fastmcp, a higher-level framework for building MCP servers. The package name mcp existed on PyPI before November 2024 for an unrelated project with a handful of downloads, so we count mcp only from November 2024 on. Many MCP servers are launched on demand with tools such as uvx, which can fetch packages each time a server starts, so one active user can produce many downloads.

From thousands to 111 million downloads a week

Weekly downloads (days 1 to 7 of each month) for mcp and fastmcp combined.

Downloads rose in 18 of the 21 month-to-month steps. The September 2026 week (62,121,122) was lower than August, in line with a 38.1% drop in total PyPI downloads between those two sample weeks.

MCP now outpaces agent frameworks

Quarterly downloads (21 sampled days) for MCP packages and the ten agent frameworks from our related study.

MCP packages made up 0.228% of every download on PyPI in the Q3 2026 sample days. Frameworks and MCP are complementary: agents built with a framework often use MCP to reach tools and data.

Nearly three in four MCP downloads use uv

Download signals for MCP packages.

The uv share fits how MCP servers are often started, by a command that sets up a fresh environment on the fly. US IP addresses accounted for 76.9% of Q3 2026 downloads, followed by Singapore at 3.7%, which mostly reflects where servers run.

What this means

MCP is how AI assistants are starting to reach live business data directly, through booking systems, product catalogs, CRMs and website content, instead of reading a cached copy of a web page. For a small business, the practical question is whether the software you already use offers an MCP connector, and whether the data it exposes, such as hours and prices, is accurate. An AI SEO consultant can help you check both before an assistant starts quoting them. An assistant that can check your real calendar is likely to rely on it over an old web page.

Your website still feeds most AI answers today, and it is the base for everything else. A free AI SEO audit shows how assistants currently describe your business, which is worth knowing before you add any connectors.

Limits

A download is not a user. One developer, one Docker build or one CI job can download the same package many times, and a single busy pipeline can move a package's numbers sharply from month to month.

Figures cover the first seven days of each month only (21 days per quarter), so they are sample totals, not full-month or full-quarter totals. Growth rates compare like-for-like sample periods.

PyPI as a whole grew from 12.1 billion downloads in the Q3 2022 sample days to 115.0 billion in the Q3 2026 sample days (9.5x), so part of every package's growth is the index growing. We report share of all PyPI downloads where it matters.

PyPI-wide downloads in the September 2026 sample week were 38.1% lower than in the August 2026 sample week (28.7 billion vs 46.4 billion), so September figures for individual packages should be read with that drop in mind.

The CI flag is only populated in this dataset from March 2024, and it only catches environments that announce themselves as CI. Automated downloads that do not set a CI variable are counted as non-CI.

Country codes come from the downloading IP address. Cloud servers sit mostly in the United States, so country splits describe where machines run, not where developers live.

On-demand launchers can download MCP packages each time a server starts, so MCP download counts may overstate the number of distinct users more than for most packages.

Only Python packages are counted. MCP is also widely built in TypeScript and other languages, which are not in PyPI logs.

Download counts do not show which MCP servers connect to which data.

Frequently asked questions

Related studies

• Downloads of the tools that read websites for AIAI Agent Framework Downloads Topped 108M in One Week
• Downloads of the tools that read websites for AIWeb-to-Text Tool Downloads Grew 71x Since 2022
• Downloads of the tools that read websites for AIYouTube Transcript Tool Downloads Rose 31x Since 2024
• Downloads of the tools that read websites for AIPDF-to-Text Downloads for AI Rose 65x in Four Years