Some time ago our team published three Python packages on PyPI: graphql-mcp-server, nife-restapi-mcp-server, nife-mcp-server. MCP servers - bridges that let AI models talk to GraphQL and REST APIs. We pushed them and got back to work. No launch post. No dashboard. Nobody checked the numbers.
This week I finally did. Last 30 days, per pepy.tech:
- graphql-mcp-server: 381 downloads - https://pypi.org/project/graphql-mcp-server/
- nife-restapi-mcp-server: 338 downloads - https://pypi.org/project/nife-restapi-mcp-server/
- nife-mcp-server: 353 downloads - https://pypi.org/project/nife-mcp-server/
1,072 installs in a month on packages nobody was marketing. The five repos behind them sit at 59 stars.
Small numbers. But a pip install is the most honest metric in software. No signup form, no email captured, no demo call. Someone had a problem, typed install, and left. You never find out who. They just keep choosing the thing you built while you are busy building the next thing.
If you maintain packages and have not looked at your numbers lately - go look. The quiet counter might be ticking.
And if you want to see what these bridges do, the code is here: https://github.com/nifetency/nife-mcp-server?utm_source=devto&utm_medium=post&utm_campaign=mcp-downloads
Top comments (1)
Pip install numbers on MCP packages are especially interesting because many developer environments pull them via ephemeral virtualenvs or automated agent runners like
uvxorpipx. When an agent workflow spins up on a worker node to query a GraphQL or REST backend, it triggers fresh installs across CI and agent runs that never show up as GitHub stars or page views.A tricky engineering challenge once download numbers climb on generic API bridges is schema discovery overhead. If an MCP server exposes an entire GraphQL schema or broad REST surface to a model on every session initialization, the tool definitions can easily chew through 10k+ tokens before the agent issues its first real query. Filtering or dynamically scoping exposed schemas based on client intent keeps both token spend and context bloat manageable as adoption grows.