Backing up the MCP ecosystem: 3% of repos gone in under a year
Written by punkpeye on .
In our first 10 months of hosting the MCP registry, we saw ~3% (325) of git repositories get deleted. That's a shame, because even if we put aside practical use cases, these were some of the first MCP servers ever published and served as an important historic record.
Therefore, partly as a preservation effort and partly forced by operational needs, we are committing to backup every open-source MCP project in the ecosystem.
Operations
Glama has a complicated MCP hosting infrastructure. The key thing to know is that we build Docker images from source code. We do this to ensure security and reproducible builds. We also rebuild Docker images every time we have security updates.
The issue we ran into is that during site-wide security rollouts, we noticed some repositories no longer existed. As a result, we couldn't apply updates to those servers. We brainstormed a few solutions, but despite its complexities and cost, keeping a backup of every git server won as the long-term solution.
Apart from just keeping our customer servers up to date, backups ensure license compliance, attribution preservation, and audit trails for enterprise users.
Early Stats
12,356 open-source authors
63,491 commits
31 GB of data (heavily skewed by people committing things like
node_modules)
Distribution by Language
Language | % |
Python | 41.0% |
TypeScript | 32.1% |
JavaScript | 18.7% |
Java | 2.7% |
Go | 1.1% |
Rust | 0.4% |
Other | 4.0% |
Implementation
Sync Frequency
Every 3, 24, or 48 hours based on popularity and update frequency. Maintainers can trigger manual syncs.
Access
Files are browsable via GitHub-like URLs:
Scope
Git trees and associated blobs only.
What's next?
Analytics: Industry adoption metrics, framework usage, and contributor insights derived from git history.
Code Search: Cross-ecosystem search to help server authors find implementation examples.
Written by punkpeye (@punkpeye)