The Fragmentation Phase: MCP’s Early Ecosystem
MCP has evolved beyond a single specification into an ecosystem with multiple registries, servers, and security tools. No canonical discovery layer exists yet. Users searching for MCP servers land on MCP registry, MCP marketplace, or GitHub lists, each showing a different slice of the ecosystem. This fragmentation is normal for phase one of ecosystem maturity. Kubernetes, Terraform providers, and REST APIs all passed through expansion and fragmentation before consolidating. MCP is currently in phase one.
The 90,000 Server Problem: Quantity vs Quality
Over 90,000 MCP servers exist across registries. The MCP zoo dataset contains 56,000 servers. The MCP marketplace lists 68,000. The MCP registry has 20,000. Yet only about 1,000 are production-ready, well-documented, and actively maintained. That is a 1.1% conversion rate. Most servers are thin API wrappers or experiment scripts written in an afternoon. 60 to 80% of the ecosystem is stale or abandoned. Only 5 to 10% receive regular updates.
Why Servers Die
Setup complexity, poor documentation, and bad repo structure kill MCP servers. If a user cannot run a server in 5 to 10 minutes, they will not use it. If they must read the code instead of documentation, they abandon the project. Onboarding friction is the biggest predictor of project failure. Repos without recognizable structure or automated tests repel contributors. Three or more red flags indicate a project headed for abandonment.
Building Servers That Last
Chowtoori presented a sustainability checklist. Five-minute onboarding is the baseline rule. CI testing serves as trust infrastructure that gives contributors confidence. Clear contribution pathways in the README tell people what to work on. Documentation must include API docs, architecture decisions, and decision records. Security posture requires proactive scanning from the start, not reactive patching. Servers with these attributes survive. Servers without them do not attract users or contributors.
The Opportunity to Shape MCP’s Future
MCP was released in 2024. Standards are still forming. Patterns are not yet set. Most developers join ecosystems after major players have already defined how things work. MCP is different. Design decisions made in the next few months will shape the ecosystem’s trajectory. Building sustainability in from the beginning raises the conversion rate from 1.1% to potentially 2%. That means fewer abandoned servers and more reliable infrastructure for everyone.
Notable Quotes
duplication is structurally inevitable in any kind of system and is not a sign of a ecosystem dysfunction essentially. Lahari Chowtoori · ▶ Watch (4:04)
we’ve got around 90,000 repos now down to thousand usable servers. That’s almost 1.1% conversion rate. Lahari Chowtoori · ▶ Watch (14:41)
the goal is not actually to have fewer MCP servers. It the goal is actually to build better ones, servers that can actually last, servers that people can depend on Lahari Chowtoori · ▶ Watch (24:36)
Key Takeaways
- MCP fragmentation signals normal phase one ecosystem maturity, not dysfunction.
- Of 90,000 MCP servers, only 1.1% are production-ready and actively maintained.
- Five-minute onboarding, CI testing, and clear documentation separate surviving servers from abandoned ones.
About the Speaker(s)
Lahari Chowtoori is an AI enthusiast and Technical Program Manager at AWS, focusing on open source, Machine Learning, and Artificial Intelligence. With a background in Data Science and Machine Learning, she is passionate about democratizing AI knowledge and fostering community collaboration.