Local vs Remote MCP Servers: Key Differences Every Developer Should Know

·1 views

Model Context Protocol (MCP) servers come in two distinct types — local (stdio) and remote (Streamable HTTP) — each suited to fundamentally different use cases. Local servers run as processes on the user's machine, accessing their filesystem, local services, and credentials, while remote servers are hosted by the developer and accessed via a URL token. The choice determines who runs the code, where secrets are stored, what resources the server can reach, and how updates are delivered. Anthropic maintains separate submission forms for each server type in its MCP directory, with the local form requiring a packaged .mcpb file — a detail that can cause developers to realize they've chosen the wrong path only after filling in the form. As a rule of thumb, if the data lives on your infrastructure, a remote server is appropriate; if it resides on the user's machine, a local server is the right choice.

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)

When Running Multiple A/B Tests Simultaneously Is Safe and When It Is Not

Experimentation teams often face the dilemma of whether to run overlapping A/B tests or isolate them, and the answer depends on how the tests interact. By default, Optimizely assigns users to experiments using independent hashing, meaning participation in one test does not influence bucketing in another, making overlap statistically safe in most cases. Interaction effects become a real risk when two tests share the same page, the same user funnel, or the same conversion metric, as one experiment can distort the results of the other in ways neither team can detect independently. A practical example is a homepage discount test running alongside a checkout redesign that removes a promo-code field, where the combination causes abnormal drop-offs that falsely make the redesign appear to underperform. Optimizely recommends using exclusion groups to enforce mutual exclusivity only when such conflict conditions are identified, since unnecessary isolation splits traffic and slows down both experiments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

FormatStack Runs All Dev Tools Client-Side, Ensuring No Data Leaves the Browser

FormatStack is a developer utility platform offering tools such as a JSON formatter, regex tester, UUID generator, Base64 encoder, and cron parser. The platform is built entirely client-side, meaning all processing occurs within the user's browser JavaScript engine without any data being transmitted to a server. According to its creator, there is no backend endpoint, no API route, and no fetch calls involved in any tool's request path. This architecture makes the privacy guarantee a structural reality rather than a policy statement, as there is no server infrastructure that could be modified to log user input. FormatStack is currently live at formatstack.tech.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Modern Startups Run on AI Agents and Automation With Minimal Human Staff

A viral post on DEV Community highlights how startup teams in 2026 are redefining what 'lean' means in the age of AI. Founders claiming small teams may actually be operating with dozens of AI agents, automated cron jobs, and multiple servers running in the background. The humorous observation points to a growing trend where automation tools do the heavy lifting once handled by larger human workforces. A single exhausted human often oversees this complex stack of digital infrastructure. The post reflects broader industry conversations about how AI is fundamentally reshaping startup operations and team structures.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AWS Guide: Enforcing AI Agent Authorization on EKS Using Cedar and Step Functions

A technical guide published on DEV Community outlines an architecture for safely governing AI-driven operations on Amazon EKS clusters. The core argument is that AI agents should propose typed remediation actions rather than execute free-form instructions, keeping authorization logic outside the model itself. Amazon Verified Permissions with Cedar policies is recommended to externalize and enforce access decisions based on deterministic infrastructure signals, not just model confidence. AWS Step Functions is proposed to manage workflow state, including human approval loops with evidence-refresh steps to prevent stale approvals from executing against an already-changed production environment. The design enforces a strict boundary where the AI proposes, trusted infrastructure verifies facts, and a separate policy layer decides whether the action is permitted.

0 comments Read more at DEV Community