MCP Protocol Update Drops Session IDs to Ease Server Scaling

A coming update to the Model Context Protocol shifts to a stateless session model, removing a key obstacle for companies trying to run MCP servers at scale.

The Model Context Protocol (MCP) — the standard that lets AI models securely connect to external data sources and services — is receiving a significant update next week that makes it easier to run at large scale, according to TechCrunch [1].

For developers and companies building agentic AI products, the change matters because the current version of MCP has made it genuinely difficult to deploy servers that handle millions of users, and that friction has slowed real-world adoption [1].

What MCP does and why session IDs became a problem

MCP is, in practical terms, the plumbing that allows a chatbot to reach into a calendar, a database, or internal tools without engineers having to build a custom connection for every integration [1]. The protocol has been one of the core building blocks of AI interoperability, but its current session-handling design creates headaches at scale [1].

Under the existing system, when an MCP client — such as Anthropic’s Claude — first connects to a server, it sends an initial handshake that includes its version and capabilities, and the server responds with its own capabilities and a session ID [1]. That session ID travels with every subsequent request so the server can recognize the ongoing conversation [1].

The problem surfaces in real deployments. A server farm handling millions of users typically sits behind a load balancer that routes each incoming request to whichever machine is free, sometimes in a different region [1]. Under the current MCP design, every machine in that farm needs to know about session IDs that other machines originally issued — work that, as TechCrunch put it, “fights the load balancer instead of working with it” [1].

The fix: a stateless approach

The updated protocol takes a looser, stateless approach to session IDs on the server side, similar to how most ordinary websites already handle requests [1]. The official specification for the new version has been public since May, though an unusually detailed explanation of the changes came from the team at Arcade on Monday [1].

The stateless model should make MCP servers easier to maintain and, in theory, cheaper to run at scale [1]. TechCrunch noted that the scaling difficulty has been “part of the reason we haven’t seen more companies ship large-scale, first-party MCP integrations despite all the hype around agentic AI this year” [1].

The update is a reminder that AI infrastructure does not always move as fast as model development itself — standards-body consensus, as TechCrunch observed, moves on its own timeline [1].


Sources

  1. TechCrunch — AI’s most important protocol is getting a little bit easier to use

This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.

A new framework from Alibaba researchers decomposes complex tasks iteratively so AI agents fetch only the tools they need, slashing context costs from ~884,000 tokens to ~1,160 per query.

Alibaba researchers have published SkillWeaver, a framework that routes AI agents to the right tools through iterative task decomposition rather than loading an entire tool library into context, reducing token consumption by over 99% compared to brute-force approaches [1].

For developers building enterprise agents that orchestrate dozens or hundreds of tools, the practical stakes are direct: fewer tokens mean lower API costs and faster responses, and the accuracy gains address a failure mode that currently makes large tool libraries nearly unusable [1].

The problem with loading every tool

Modern large language model (LLM) agents are increasingly integrated with massive tool ecosystems — including systems built on the Model Context Protocol (MCP), a standard for connecting agents to external services [1]. When an agent needs to handle a multi-step request like “Download the dataset, transform it, and create visual reports,” exposing the full tool library to the LLM to find the right tools is, as the researchers describe it, “highly inefficient” and “quickly overwhelms context limits” [1].

Most existing frameworks treat tool selection as a one-shot, single-skill problem, which breaks down when real business queries require sequencing multiple tools in a compatible chain [1].

How SkillWeaver works

SkillWeaver structures tool selection across three stages: Decompose, Retrieve, and Compose [1]. First, an LLM breaks a complex user query into atomic sub-tasks, each requiring exactly one tool [1]. An embedding model then searches the tool library to produce a shortlist of candidates for each sub-task [1]. Finally, a planner checks inter-tool compatibility — whether the output of one tool feeds cleanly into the next — and assembles the result as a Directed Acyclic Graph (DAG), allowing independent steps to run in parallel [1].

The central innovation is a feedback mechanism the team calls Skill-Aware Decomposition (SAD) [1]. Rather than decomposing a task once and searching, SAD has the LLM draft an initial plan, runs a preliminary search to surface loosely matching tools, and feeds those results back to the LLM so it can rewrite its decomposition using the actual vocabulary of the available tools [1]. The authors shared the prompt templates in their paper, and the technique can be reproduced with standard libraries such as LangChain, LlamaIndex, or plain Python [1].

Benchmark results

The team evaluated SkillWeaver on CompSkillBench, a custom benchmark of 300 multi-step queries run against a library of 2,209 real-world skills drawn from the public MCP ecosystem across 24 categories including cloud infrastructure, finance, and databases [1].

Without SAD, a 7-billion-parameter Qwen2.5-7B-Instruct model correctly predicted the number of required steps only 51.0% of the time; activating SAD pushed that to 67.7%, and using the larger Qwen-Max model brought it to 92% [1]. On hard tasks requiring four to five tools, SAD improved accuracy by 50% [1].

One counterintuitive finding: a 14-billion-parameter model performed worse than the 7B model in the unguided setup because it tended to over-decompose tasks into unnecessary micro-steps [1]. SAD corrected this by anchoring the larger model to the actual tool vocabulary, suggesting that tool alignment can matter more than raw model size [1].

The token savings were stark. The brute-force baseline — feeding all tool names into a large model’s prompt — consumed an estimated 884,000 tokens per query and still only retrieved the correct tool category 21.1% of the time [1]. SkillWeaver’s targeted approach brought that figure down to roughly 1,160 tokens per query, a 99.9% reduction, while substantially outperforming on accuracy [1]. A ReAct-style agent baseline achieved 0% decomposition accuracy, collapsing multi-step plans into isolated actions [1].

On retrieval infrastructure, the researchers used the open-source all-MiniLM-L6-v2 embedding model with a FAISS index [1]. Indexing all 2,209 skills took 15 seconds, and retrieval adds under 15 milliseconds of latency per query [1]. Swapping in a stronger encoder (BGE-base-en-v1.5) improved accuracy without any fine-tuning [1].

Limitations and what’s not included

The framework currently has no error-recovery mechanism [1]. If a tool call fails mid-chain, the entire execution breaks, and the authors acknowledge that practitioners will need to build their own retry and fallback logic on top of the compose stage [1].

The source code has not yet been released, though the researchers note the system was built entirely from off-the-shelf components and the SAD prompt templates are available in the paper [1].


Sources

  1. VentureBeat — New Alibaba AI framework skips loading every tool, cutting agent token use 99%

This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.

AWS Context automatically builds and updates a knowledge graph from existing enterprise data, aiming to eliminate manual re-curation as agents learn from usage.

Amazon Web Services on Wednesday announced a trio of products it is positioning as a context intelligence stack for AI agents, with a self-learning knowledge graph service called AWS Context at the center. [1]

The context layer — the architectural tier that sits between enterprise data stores and AI agents, giving those agents the structured knowledge they need to answer questions accurately — has become a contested product category, with offerings from Snowflake, Microsoft, Redis, and Pinecone already in the market. [1] AWS is entering that space with a different premise: the graph should update itself based on how agents actually use it, rather than requiring human re-curation. [1]

“This service automatically builds a knowledge graph from all your existing data,” said Swami Sivasubramanian, vice president of Agentic AI at AWS, during his AWS Summit NYC keynote. [1] The service maps relationships across existing data — what tables exist, what columns mean, how sources relate, and which sources are authoritative — and combines semantic search with graph-level reasoning to make that context available to agents at runtime. [1]

“The knowledge graph improves itself over time as it learns which sources produce correct results and which parts get used,” Sivasubramanian said. [1] Data stewards can review inferred relationships, promote them to production, and attach business definitions and usage rules through the AWS Management Console. [1]

On the security and governance side, every query inherits the calling user’s AWS Identity and Access Management (IAM) and Lake Formation permissions, making agent data access auditable by identity through controls enterprises already rely on. [1] All metadata is published in Apache Iceberg format to Amazon S3 Tables and is queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine, with no proprietary APIs required. [1]

AWS Context is not a standalone product. It synthesizes inputs from two additional services announced alongside it. [1] Amazon S3 Annotations — now generally available — lets users attach rich business context directly to individual S3 objects at the storage layer. [1] A preview of skill assets in AWS Glue Data Catalog, the company’s metadata management service, allows domain knowledge such as runbooks, query patterns, and usage rules to be linked to data assets across an organization’s data estate. [1] AWS Context then pulls both layers together into the knowledge graph that agents query at runtime. [1]

Agents can query the graph through agentic search application programming interfaces (APIs) and Model Context Protocol (MCP) tools across Bedrock AgentCore, Amazon Elastic Kubernetes Service (EKS), or any MCP-compatible framework. [1] Third-party catalog connections are also supported, so context from systems outside AWS can be incorporated into the same graph. [1]

AWS’s pitch to enterprises already running S3, Glue, and Lake Formation is that AWS Context extends an existing identity model with no data movement required. [1] Competitors in the space include Snowflake, which announced its Horizon Context and Cortex Sense services earlier this month; Microsoft, which provides context via its Fabric IQ platform using a semantic ontology for data; Redis, which has developed a context platform optimized for retrieval; and Pinecone, whose Nexus offering compiles enterprise data into task-specific artifacts before agents query them. [1]

Holger Mueller, vice president and principal analyst at Constellation Research, told VentureBeat that “context makes agents more powerful and as the whole world is building agents, every agentic platform vendor needs a context capability.” [1] Mueller also flagged a potential limitation shared across the category: “The concern — as with all context offerings — is going to be performance, especially for transactional data,” he said. [1]


Sources

  1. VentureBeat — AWS enters the context layer race with a graph that learns from agents, not manual curation

This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.