MCP Protocol Drops Session IDs, Shifts State Visibility to the Model

·1 views

The July 2026 revision of the Model Context Protocol eliminated the initialize handshake and Mcp-Session-Id header, making every request self-describing by embedding protocol version and client info directly in each call. Previously, session state was negotiated once at connection time and held invisibly inside a server process, entirely out of reach of the AI model doing the reasoning. Under the new design, servers that need to track state across calls return an explicit handle — such as a basket_id — as a tool result, which the model then passes back as a normal argument on subsequent calls. This means stateful workflows are still supported, but the state is now visible to, and reasoned over by, the model rather than hidden in the transport layer. A notable side effect is that stateless requests can be routed to any server instance, making MCP servers compatible with serverless infrastructure and standard load balancers.

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)

GopherKind bridges Nostr and Gopher protocol, turning every public key into a gopherhole

A developer has built GopherKind, an open-source tool that maps Nostr public keys to Gopher protocol content, making every Nostr account automatically accessible as a gopherhole. The project converts a user's Nostr profile, notes, replies, articles, and social graph into standard Gopher menus and text files. Authors can also publish dedicated, path-addressed pages using a new proposed event type called kind 31436, defined in a draft NIP submitted as PR #2429 to the nostr-protocol repository. The tool aims to solve Gopher's longstanding lack of authorship and content persistence by anchoring documents to cryptographic identities via Nostr's signed events. Content published through GopherKind is accessible via Gopher, Gemini, the web, and terminal without requiring readers to hold a Nostr account.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

German AI Adoption Tops 51% as Language Gaps Complicate Agentic AI Deals

A March 2026 Salesforce and Deutscher Mittelstands-Bund report found that 51.2% of German mid-sized companies are now using or testing AI, up from 33.1% the previous year. The German-speaking market lacks a precise equivalent for the English adjective 'agentic', relying instead on the noun 'KI-Agent', which conveys no information about a system's level of autonomy. This vocabulary gap creates real procurement problems, as contracts for 'einen KI-Agenten' rarely specify how much decision-making authority is delegated, leaving room for supplier misinterpretation. Five terms are currently used interchangeably in the German market — including 'Agentische KI', 'Autonome KI-Systeme', and 'KI-Mitarbeiter' — despite carrying meaningfully different implications. The term 'KI-Mitarbeiter', or AI employee, poses additional legal risk under EU transparency rules taking effect on 2 August 2026, which restrict presenting AI systems as human to customers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer finds 6 published pages missing from tracking scripts due to silent ledger bug

A developer maintaining a plain-text ledger of published URLs discovered that six entries had gone untracked by all monitoring scripts for eleven days. The pages were stored in a separate table within the same file, causing a status-based filter to skip them entirely without raising any error. Unlike a misclassified or missing value, the omitted rows produced no alerts because nothing in the system was designed to count what the filter ignored. The developer fixed the blind spot by writing a secondary script that counts all URLs in the ledger regardless of row structure, then compares that total against the status-filtered count. The incident highlighted four tiers of data-tracking failure, with silent omission and correct-but-misdirected records being the hardest to detect.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How a Unique DB Constraint Prevents Double-Selling Seats at Scale

A software engineer who built booking systems for events hosting up to 20,000 spectators has shared the architecture behind reliable seat reservation at high concurrency. The core problem is the 'lost update' — two users passing an availability check simultaneously, both being assigned the same seat. The solution separates clicking from selling: a click creates a short-lived hold stored in Redis via an atomic Lua script, while only a completed checkout writes to the database. Redis handles the high-churn, expiring holds without polluting the primary database, but the true guarantee is a PostgreSQL UNIQUE constraint on the event and seat combination. A reference implementation with a 200-concurrent-request stress test has been published on GitHub to demonstrate the approach.

0 comments Read more at DEV Community