The Integration of Arcade.dev and LangSmith Fleet for AI Agents

LangSmith Fleet, a platform designed to facilitate the creation, use, and sharing of AI agents for daily tasks, has announced a new strategic partnership with Arcade.dev. This collaboration introduces Arcade's extensive tool library into the Fleet ecosystem, providing a unified and secure gateway for accessing a wide range of applications. Arcade.dev positions itself as an MCP (Multi-tool Coordination Protocol) runtime for production agents, focusing on secure agent authorization, tool reliability, and governance.

The primary goal of this integration is to simplify the management of agents operating across multiple tools. Fleet agents are designed to interact autonomously with various applications, such as pulling data from Salesforce, updating pages in Notion, or sharing results in Slack. Reliable and secure access to these tools is critical, and Arcade.dev's MCP gateway aims to solve this challenge by offering a single connection point for over 7,500 agent-optimized tools.

Centralized Gateways: Efficiency and Control for AI Operations

Managing connections to external services represents a significant complexity for AI agent deployments. Centralized gateways are emerging as an effective pattern to streamline access. Similar to gateways for Large Language Model providers, which centralize access and credentials, a comparable approach is crucial for tool management. Each new tool integration involves specific authentication flows, API quirks, and ongoing maintenance burdens. Multiplying this complexity by the growing number of tools used by a team, the Total Cost of Ownership (TCO) and engineering effort can rapidly escalate.

Arcade.dev's MCP Gateway provides a single access point for agents. Organizations can connect their Arcade account in Fleet, select their desired gateway, and within minutes, agents gain access to a wide array of applications such as Salesforce, Asana, and Zendesk. This architecture allows for the creation of a single gateway for the entire organization or tailored gateways per team or use case, enabling users to connect with their own credentials and access only the tools relevant to their work, without adding to the engineering team's backlog.

Agent-Optimized Tools: Beyond Simple API Wrappers

In the current landscape, numerous MCP servers often merely "wrap" existing REST APIs within the MCP protocol. While this can facilitate standardized tool discovery, it does not alter the underlying functionality of the API. This distinction is crucial when agents are the callers of these APIs. Traditional APIs were designed for human programmers, who decide which endpoint to call and how to structure the request. They expose large surfaces with many endpoints and parameter combinations, describing data shapes rather than intent. They expect structured inputs and return raw HTTP errors when something goes wrong. An agent operating from natural language context must navigate all of this complexity, risking "hallucinated" parameters, poor tool selection, or wasted tokens cycling through irrelevant endpoints.

Arcade.dev addresses this issue by offering MCP tools specifically designed for agents. These tools are narrowed to what agents actually need to do, rather than exposing the full API surface. Each tool follows consistent structural patterns, and tool descriptions are written to optimize selection and invocation by Large Language Models. This approach significantly improves agents' ability to choose and use tools effectively, reducing errors and increasing reliability.

Security and Control for Enterprise AI Agent Deployments

Security is a fundamental aspect of adopting AI agents in enterprise contexts, especially for those evaluating on-premise deployments or hybrid environments where data sovereignty and control are paramount. LangSmith Fleet and Arcade.dev collaborate to manage tool authentication and authorization for your agents. Arcade handles per-user, session-scoped authorization, enforcing the principle of least privilege at runtime. This means that each action performed by an agent inherits the permissions of the specific user the agent is acting for, ensuring that different individuals have different levels of access to downstream systems.

Fleet provides the configuration for how credentials flow into Arcade. Agents configured as "Assistants" pass each user's own credentials when tool calls are made, so actions reflect that user's permissions in the downstream system. Conversely, agents configured as "Claws" use a fixed set of credentials shared across all users, which is useful when the agent is acting on behalf of a team or service rather than an individual. This granularity in access control is essential for enterprises requiring strict compliance and a robust security architecture for their AI workloads. For those evaluating on-premise deployments, AI-RADAR offers analytical frameworks on /llm-onpremise to assess the trade-offs between control, security, and operational costs.