Benchmark Reveals MCP Servers Load 200K Tokens of Overhead Before First User Message

·1 views

A developer benchmarked 10 popular MCP servers from the official registry and measured token consumption injected into the context window before any user input. Across all 10 servers, the combined overhead reached over 200,000 tokens, with Google Drive alone accounting for 47,293 tokens due to its 31 deeply nested tool schemas. At Claude 3.5 Sonnet pricing, a developer running 10 servers across 20 daily conversations could spend roughly $216 per month — exceeding the cost of a Claude Pro subscription. The analysis found that inputSchema properties represent the largest cost driver, making up 42% of total token usage, while even tool call results carry roughly 32% JSON structural overhead. The author released an open-source CLI proxy called mcptoon aimed at reducing this overhead through schema caching and compression.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Tailscale Workaround Saves Healthcare IT After Remote VPN Server Fails Post-Patch

A healthcare organization faced a major IT crisis when its VPN server, located 250 kilometres away, failed to come back online after a routine restart and patch update. The outage also knocked out access to the server's integrated Dell Remote Access Controller (IDRAC), eliminating all remote management options and making a costly physical trip the only apparent solution. A single IT administrator resolved the crisis by deploying Tailscale, a peer-to-peer VPN tool, on a machine within the same network using Microsoft's live response capability, restoring access without on-site travel. The root cause was a VPN virtual machine that failed to auto-start, a symptom of deeper architectural weaknesses including no redundancy and reliance on a single point of failure. The incident highlights how underfunded IT infrastructure in critical sectors like healthcare can turn routine maintenance into operational emergencies with serious consequences for patient care and service continuity.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build a Custom Kubernetes Scheduler Using Simple Bash Scripts

A hands-on lab guide demonstrates how Kubernetes scheduling works by walking through the creation of a custom scheduler from scratch using bash scripts. The Kubernetes scheduler follows three phases — filtering, scoring, and binding — to intelligently assign pods to nodes based on CPU, memory, and other constraints. The lab shows how to configure an nginx pod to use a custom scheduler named 'my-scheduler,' which causes the pod to remain in a Pending state until that scheduler is active in the cluster. A simple open-source bash script is then used to mimic core scheduler behavior, including querying available nodes, identifying waiting pods, and making binding requests to the Kubernetes API server. While production schedulers are typically built in Golang, this approach offers a clear, accessible way for DevOps engineers to understand the underlying scheduling workflow.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developers Find Go Faster to Compile and Easier to Set Up Than Swift

Developers switching from Swift to Go report significantly faster compile times, attributed to Go's single-pass compilation model versus Swift's multi-pass architecture that performs extensive type-checking and whole-module optimization. Swift's thorough compiler design improves runtime performance but creates longer wait times and heavier resource usage during the build process. Go's simpler type system and opinionated design further reduce compilation overhead, resulting in quicker feedback loops for developers. Beyond speed, Go's built-in tooling lowers onboarding friction, allowing developers to get projects running with minimal setup compared to Swift's dependency-heavy project configuration. While Swift may still be preferable for performance-critical applications requiring runtime optimization, Go is seen as the stronger choice for rapid iteration and everyday development workflows.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Beginner Documents Two-Week Journey Learning Git, GitHub, and SSH Workflows

A student at Luxe Dev HQ completed a two-week introductory data analysis course focused on Git and GitHub fundamentals. The learner discovered that Git and GitHub serve distinct purposes — Git is a version control tool for tracking changes, while GitHub is an online platform for storing and collaborating on repositories. Using Windows PowerShell, the student set up SSH authentication, created a local directory, and linked it to a remote GitHub repository. The process involved core Git commands including git status, git add, git commit, and git push to transfer files from a local folder to GitHub. Despite spending several hours troubleshooting SSH configuration, the student successfully completed the workflow with assistance from instructors.

0 comments Read more at DEV Community