How MCP Protocol Uses JSON-RPC Handshakes to Connect AI Models to External Tools

·1 views

The Model Context Protocol (MCP) is an open standard designed to bridge the gap between stateless large language models and external tools, APIs, and data sources. At its core, MCP relies on a JSON-RPC 2.0 handshake in which a client first sends an initialize request and the server responds with its capabilities before a tools/list call retrieves structured definitions of available tools. Each tool definition includes its name, description, and a JSON Schema for inputs, enabling AI models to understand not just what tools exist but precisely how to invoke them. Developers can implement a functional MCP server in approximately 50 lines of Go using the official go-sdk, exposing tools such as a compound interest calculator with strictly typed input schemas. By standardizing discovery and integration, MCP aims to replace fragile, custom-built AI connectors with a composable, model-agnostic architecture.

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)

Baidu's Unlimited-OCR Parses 40-Page PDFs in One Pass With Flat Memory Use

Baidu has released Unlimited-OCR, an open-source model capable of processing multi-page PDFs in a single forward pass without the memory growth that typically plagues OCR pipelines. The model replaces standard attention layers with a mechanism called Reference Sliding Window Attention (R-SWA), which maintains a fixed-size token cache instead of an ever-growing one, keeping memory and latency constant across long documents. With 3 billion total parameters but only 500 million activated via a Mixture-of-Experts architecture, the model runs closer to the cost of a 500M model at inference time. It scored 93.23 on OmniDocBench v1.5, roughly 6.2 points above the DeepSeek-OCR baseline it was trained from, suggesting the efficiency gains did not come at the cost of accuracy. The model is released under an MIT license and requires an NVIDIA GPU with at least 12 GB of VRAM for practical use.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Engineer Streams USB Mouse Over WiFi Using Netevent and Systemd

A developer without a USB-C adapter devised a way to use a USB mouse connected to one machine on a separate laptop over the network. The solution uses netevent, a small Linux tool that captures raw input events from a device and streams them via a pipe to recreate a virtual mouse on the receiving machine. An initial SSH-based one-liner worked but introduced security and reliability concerns, prompting a more robust setup. The author replaced it with a systemd socket-activated service bound exclusively to a Tailscale network interface, ensuring encrypted, authenticated access with no LAN exposure. Additional hardening measures included limiting simultaneous connections to one and using a short-lived dynamic user with only the input group privilege, eliminating the need for root access.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

JavaScript Function Types: Key Differences in Hoisting, This, and Usage

JavaScript offers multiple ways to define functions, including declarations, expressions, arrow functions, object methods, constructors, classes, generators, async functions, and IIFEs. Each type differs in important behaviors such as hoisting, how the 'this' keyword is bound, access to the 'arguments' object, and whether the function can be invoked with 'new'. For example, function declarations are fully hoisted and callable before their definition, while arrow functions are not hoisted and inherit 'this' lexically from their surrounding scope. Arrow functions used as object methods can cause bugs because they do not bind 'this' to the owning object, often returning undefined instead. Understanding these distinctions is essential for writing predictable JavaScript and is a common focus area in technical interviews.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

PostgreSQL Cursors Slash Export Times by Replacing Random Reads with Sequential Scans

A data pipeline exporting hundreds of gigabytes from PostgreSQL in 2,000-row batches was running far slower than expected, with the bottleneck traced to how the database read data from disk. Using a WHERE id > X LIMIT 2000 pattern caused Postgres to perform up to 2,000 random page lookups per batch via a B-tree index, forcing the disk head to constantly seek across different locations. Switching to a server-side cursor allowed Postgres to scan the heap sequentially, reading each 8KB page exactly once and enabling the OS to prefetch ahead predictably. This change required no schema modifications — only a different query pattern — yet dramatically reduced I/O overhead at scale. However, cursors hold an open transaction for their entire duration, which can block vacuum and inflate write-ahead logs on very large or long-running exports.

0 comments Read more at DEV Community