March 2026 Summaries

Arcade.dev has been recognized in the 2026 Enterprise Tech 30 award list, highlighting its role as a leader in advancing enterprise AI alongside notable companies like Anthropic and OpenAI. The company started in 2024 by addressing the challenge of enabling agents to perform actions safely, focusing on runtime authorization across multiple users with audit trails. By 2025, Arcade capitalized on the rise of MCP, launching the first MCP runtime and collaborating with Anthropic on the URL Elicitation spec to standardize OAuth handling, leading to partnerships with major enterprises. In 2026, the landscape shifted as long-running, tool-calling agents became mainstream, prompting Arcade to tackle complex issues of trust, reliability, and governance in agentic workflows. With agents now operating at scale across various platforms on behalf of users, Arcade's early focus on solving these challenges has positioned it at the forefront of enterprise AI, as the market increasingly prioritizes safe and effective agent deployment.

AI teams are advancing beyond basic chat interfaces, implementing agents capable of interacting with internal systems and executing real actions, which necessitates enhanced controls over authorization, governance, and security. The partnership between Akto and Arcade.dev addresses these needs by combining Arcade's MCP runtime, which enables scalable deployment of multi-user agents with built-in authorization and governance, with Akto's security features that provide visibility, monitoring, and control over agent behavior. This collaboration ensures that AI agents can act safely, with Akto offering security testing, risk assessment, and real-time evaluation of tool interactions, while Arcade manages the infrastructure and execution pipeline. The integration, utilizing Arcade's Contextual Access hooks, allows Akto to evaluate tool inputs and outputs dynamically, preventing unsafe actions and ensuring secure interactions without adding operational complexity. This partnership is pivotal for teams seeking to deploy agents that are both functional and secure, bridging the gap between innovative agent capabilities and robust security measures in AI systems.

Melissa Herrera, a Senior Developer Advocate at Temporal, discusses the challenges and solutions associated with complex agentic workflows, particularly the issue of error compounding in multi-step processes. She argues that treating tools as workflows rather than standalone functions can mitigate these issues, drawing parallels to distributed systems engineering solutions that have existed prior to the AI boom. Temporal's Durable Execution model offers a framework where failures in workflows are recoverable through checkpoints, automatic retries, and self-healing. This approach transforms potentially fragile, error-prone workflows into robust, dependable systems, similar to video game savepoints that prevent players from losing all progress. The architecture of Temporal involves Workflows, Activities, and Workers, enabling deterministic orchestration while handling non-deterministic tasks like API calls. Herrera emphasizes the relevance of these distributed system patterns to AI agents, which face similar challenges due to their dependence on external APIs and complex chains of operations. Temporal's primitives, including Signals and Queries, enhance interaction with long-running processes by allowing human intervention and state inspection without disrupting the workflow. Herrera sees the potential for agents to be treated as tools within this framework, promoting a modular, resilient approach to building complex agentic systems. She highlights that the issues AI engineers face today echo those from earlier distributed systems architectures, reinforcing the importance of durability in modern AI applications.

ToolBench is a new benchmarking system designed to evaluate MCP servers on their readiness for production use, focusing on four key dimensions: definition quality, protocol compliance, security, and supportability. It grades servers based on how well they meet these criteria, with the evaluation framework informed by real-world deployments and expert tools like Arcade's Agentic Tool Patterns and Nate Barbettini's MCP Debugger. Currently, ToolBench has indexed 41,902 servers and analyzed 218,422 tools, with only 0.5% achieving an A grade or higher, highlighting widespread quality issues such as missing descriptions and inadequate error handling guidance. The goal of ToolBench is to improve the reliability of tools used in production by providing a transparent scoring system that developers can use to audit and improve their MCP servers, fostering a more robust ecosystem. The benchmark aims to elevate the standard of MCP tools, thereby enhancing the performance of agents in production settings.

Jack Herrington, known as the Blue Collar Coder on YouTube, is a key contributor to TanStack and is developing TanStack AI, a type-safe, provider-agnostic software development kit (SDK) designed to integrate AI into web applications. Emphasizing an open-source, opinionated approach without vendor lock-in, TanStack AI builds on the philosophy of popular TanStack libraries by supporting adapters for various AI and streaming services and integrating smoothly with existing TypeScript stacks. A notable feature, Code Mode, allows agents to efficiently orchestrate tool calls by generating and executing TypeScript programs, reducing context usage and enhancing performance. TanStack AI supports JSON Schema tool definitions, making integration straightforward for systems using the Model Context Protocol (MCP) and enabling cross-language interoperability. By facilitating seamless tool orchestration and offering built-in approval workflows, TanStack AI empowers developers to create dynamic, user-responsive applications, where capabilities can adapt to user needs, shifting focus from predefined features to flexible functionalities.

The text discusses the development and advantages of mcpx, a command-line tool that facilitates interactions between coding agents and MCP servers by using a familiar shell interface. It highlights the limitations of integrating coding agents directly with raw APIs due to their complexity and inefficiencies in token usage, and contrasts this with the streamlined, task-focused approach of MCP, which offers a standardized protocol for handling capabilities, schemas, and authentication. mcpx simplifies the process by allowing agents to search, inspect, and execute tasks without managing persistent connections or bloating context windows with unnecessary tool schemas. The tool is particularly suited for single-user, single-machine scenarios, offering efficient token use and reducing complexity while maintaining support for the advancing MCP protocol features. This approach emphasizes the importance of good tooling, akin to curl for HTTP or jq for JSON, to ensure agents can effectively utilize MCP through a user-friendly CLI. mcpx also includes semantic search capabilities and full support for the evolving MCP specifications, making it a valuable asset for developers integrating various services without needing to configure each one individually.

Arcade.dev has introduced five new MCP servers that enable agents to fully interact with Microsoft Office files, including Word, Excel, PowerPoint, OneDrive, and SharePoint, overcoming previous limitations in handling binary file formats and complex session management. These tools provide full read and write capabilities, allowing agents to perform tasks such as updating spreadsheets, creating presentations, and managing files within OneDrive and SharePoint, all while adhering to Microsoft Graph's authentication and concurrency requirements. By abstracting the intricacies of the Graph API, these MCP servers streamline the automation of document-heavy workflows, significantly enhancing productivity for enterprise users. The new tools are designed with an emphasis on simplicity and effectiveness, enabling agents to execute complex operations like generating summaries, drafting proposals, and updating reports effortlessly.

Arcade's Salesforce MCP server has significantly expanded its capabilities to streamline CRM workflows by integrating 17 tools that automate the entire sales process, from lead generation to deal closure. Initially limited to searching accounts and creating contacts, the server now allows AI agents to perform a variety of tasks, including creating, updating, and converting leads, managing opportunities, scheduling tasks, and logging calls, all of which traditionally required manual input from sales representatives. This enhancement addresses the challenges posed by Salesforce's complex API, such as picklist validation and SOAP-based lead conversion, by offering tools that handle these intricacies, thereby reducing the need for agents to engage in repetitive data entry. The server supports concurrent request limits and pagination to efficiently manage API calls, ensuring seamless integration with Salesforce. By automating routine CRM tasks, Arcade empowers sales teams to focus more on selling rather than on administrative tasks, thus enhancing productivity and user experience.

Arcade's Microsoft Office 365 MCP servers provide AI agents with the capability to perform read and write operations across the full range of Microsoft Office applications, including Word, Excel, PowerPoint, OneDrive, and SharePoint, overcoming previous limitations posed by complex OAuth authentication and file format challenges. This suite of over 30 production-grade tools allows agents to handle binary file formats, manage sessions, deal with concurrent edits, and navigate the intricacies of Microsoft Graph, enabling seamless updates and edits to Office files without manual intervention. The integration simplifies traditionally labor-intensive tasks, such as updating spreadsheets, drafting proposals, or summarizing presentations, by allowing agents to automate these processes, thus significantly enhancing productivity in document-heavy workflows. Through Arcade's infrastructure, users can authenticate using their own Microsoft accounts, while the system manages tokens, scopes, and lifecycle, making it easier for developers to build agent-based solutions without delving into the complexities of Microsoft API endpoints.

Michael Hablich, a seasoned developer tools expert at Google, is revolutionizing coding agents with the introduction of Chrome DevTools MCP Server, which offers debugging capabilities akin to those provided by Chrome DevTools to web developers. This platform allows coding agents to perform detailed debugging and automation tasks by enabling them to read console messages, inspect network requests, and handle performance profiling through the Chrome DevTools Protocol. The system integrates with coding agents like Gemini CLI and Claude Code, enhancing their ability to autonomously identify and fix runtime issues, as exemplified by its deployment at CyberAgent, a major Japanese tech company. The evolution of this server, which initially faced challenges due to complexity, now focuses on providing low-level tools with clear interfaces for agents to utilize effectively. With features such as memory debugging and a command-line interface, Chrome DevTools MCP is making strides in simplifying complex debugging processes and expanding the capabilities of coding agents in web development.

Arcade.dev provides an automated audit logging system for AI agent runtimes that captures every administrative action across the infrastructure, offering a detailed, immutable record of changes including who made them, what was altered, how, and when. This system is enabled by default, requiring no additional configuration, and includes a filterable dashboard and a REST API to facilitate integration with existing compliance and reporting tools. Audit logs are essential for maintaining production deployments by providing clear change management evidence, which is crucial for compliance with frameworks like SOC 2 and ISO 27001. The logs ensure that when issues arise, such as workflow disruptions due to API key rotations or configuration changes, teams can quickly identify the cause and maintain momentum without lengthy investigations. The audit logs retain records for 30 days, and while currently focused on administrative actions, future updates aim to cover individual tool executions as well.

Arcade.dev's partnership with Daytona offers a comprehensive toolkit designed to streamline multi-system orchestration for AI agents, enhancing their ability to perform complex workflows across various platforms like Linear, GitHub, Slack, and Google Docs. The Daytona toolkit, integrated with Arcade's MCP runtime, comprises 46 tools that facilitate sandbox management, code execution, file system operations, and Git actions, making it ideal for long-running, autonomous agent workflows. Key features include name-based sandbox resolution, fuzzy matching, structured error recovery, and LLM-optimized tool descriptions that ensure reliable and efficient execution. The toolkit also introduces seamless GitHub OAuth integration, eliminating the need for manual token management and securely enabling Git operations. This setup allows for advanced autonomous development processes such as intelligent continuous integration, automated regression bisecting, and continuous refactoring, all conducted within isolated sandboxes to maintain security and efficiency.

Sam Morrow and his team at GitHub developed the most-used remote MCP server in the world, despite initially building it as an internal side project. The server gained viral attention when it coincided with the launch of VS Code's Agent mode, leading to widespread adoption and challenges in user interaction. The team discovered that a granular approach to GitHub's API wasn't effective as agents struggled with too many tools, highlighting the need to design toolsets that match common workflow patterns. They emphasized the importance of semantic grouping, meaningful error messaging, and filtering tools based on authentication token scopes. Sam shared insights on designing MCP servers, advocating for prompt-first tool design, composable toolsets, and dynamic tool selection to enhance user and agent interactions. He also stressed the significance of annotations in MCP tools for better decision-making and user protection. The GitHub MCP server continues to evolve with community input, offering new features and encouraging engagement through forums and working groups.

In an analysis comparing the token consumption of two MCP toolkits, Arcade and Composio, within the Attio CRM sandbox environment, it was found that Arcade was significantly more efficient, consuming 7,426 tokens across eight CRM queries compared to Composio's 747,083 tokens, a difference of over 100 times. This disparity impacts infrastructure costs and agent reliability, as shown by detailed benchmarking involving identical queries run through both toolkits on the same CRM data, which included 50 companies, 100 contacts, and 50 deals. The efficiency gap is attributed to Arcade's ability to specify needed fields and streamline metadata, whereas Composio outputs full records and detailed API metadata, resulting in higher token usage. For agents performing complex workflows, Arcade's lower token usage offers substantial cost savings, with potential annual savings reaching up to $2 million at enterprise scales when considering the higher token prices of different models. This efficiency is crucial, as research indicates that long-context inputs can degrade performance, highlighting the importance of optimizing token usage in large-scale operations.