Back to list
Product LaunchAI AgentsModel Context ProtocolDeveloper Tools

Scalekit Launches AgentKit: An Enterprise MCP Gateway Bringing Delegated Authentication to AI Agents

Scalekit has launched AgentKit, an enterprise-grade Model Context Protocol (MCP) gateway designed to handle authentication, tool calling, and permissions for production AI agents. As autonomous agents advance from experimental prototypes to mission-critical business workflows, software engineering teams encounter complex operational hurdles surrounding user identity management, credential security, and tool bloat. AgentKit bridges this gap by offering a managed infrastructure layer connecting agents to over 600 applications and 30,000 tools. Through delegated per-user OAuth tokens, isolated credential vaults, Virtual MCP Servers, and real-time audit logging, AgentKit guarantees that agents act strictly on behalf of verified users within authorized boundaries. By abstracting away integration plumbing and compliance challenges, AgentKit empowers developers to ship safe, scalable, and audit-ready agentic software into enterprise production environments.

Product Hunt

Key Takeaways

  • Unified MCP Gateway Infrastructure: AgentKit connects autonomous AI agents to more than 600 SaaS platforms and over 30,000 distinct tools through a single Model Context Protocol gateway.
  • Per-User Delegated Authentication: Actions performed by an agent run strictly on behalf of the interacting user, leveraging individual OAuth flows, managed token lifecycles, and isolated credential vaults.
  • Elimination of Tool Bloat: With Virtual MCP Servers, developers can assemble and expose only the specific tool subsets required for a given agent role, reducing context overhead and prompt pollution.
  • Event-Driven Triggers and Custom Integrations: The platform supports real-time event triggers to initiate agent execution instantly, alongside custom connectors for proprietary internal APIs and microservices.
  • Enterprise Security and Compliance: Scalekit provides complete execution audit logs, Bring Your Own Key (BYOK) encryption, IP allowlisting, and enterprise-grade posture tailored for SOC 2 and HIPAA environments.

In-Depth Analysis

The Production Bottleneck: Agent Identity and Scoped Tool Execution

The AI ecosystem has experienced a massive paradigm shift from conversational chatbots to autonomous agents capable of executing multi-step actions across software environments. However, moving agents from local sandboxes into enterprise production exposes severe infrastructure challenges, primarily centered on identity, permissions, and credential governance. Traditional API integration patterns often rely on monolithic, shared service accounts. In an agentic environment, granting a shared bot high-level system credentials introduces catastrophic security vulnerabilities, violating the fundamental principle of least privilege and making forensic auditing nearly impossible.

AgentKit solves this architectural dilemma by introducing a delegated access model powered by per-user authentication. When an AI agent performs an operation—such as sending an email via Gmail, creating a pull request on GitHub, or updating an opportunity in Salesforce—the action is executed under the authenticated identity of the specific human requesting the task. Scalekit manages the entire OAuth lifecycle, token issuance, and token refresh mechanisms outside the agent's application code and language model context window. Credentials reside inside secure vaults with AES-256 encryption at rest, preventing sensitive keys from ever leaking into prompts, system logs, or third-party inference providers.

Virtual MCP Servers and Custom Connectors: Mitigating Context Bloat

As organizations connect AI agents to comprehensive enterprise ecosystems, they encounter the phenomenon of tool bloat. Modern frontier models suffer degraded reasoning performance, increased inference latency, and excessive token costs when hundreds of tool definitions are simultaneously injected into their system context. Furthermore, exposing an unconstrained set of external capabilities increases the risk of hallucinated tool invocations or unintended actions.

To address this technical hurdle, AgentKit incorporates Virtual MCP Servers. Instead of exposing thousands of endpoints to an agent simultaneously, developers can construct modular virtual endpoints that selectively expose only the precise tools necessary for a specific job function. For example, a customer support agent can be limited strictly to Zendesk search and Slack notification schemas, while an internal engineering agent accesses GitHub and Jira. In addition to pre-configured connectors for over 600 popular applications, AgentKit provides custom connector capabilities. Engineering teams can register proprietary internal REST endpoints, gRPC services, and private microservices into the gateway, automatically wrapping them with enterprise authentication, role-based access control, and telemetry.

Operational Infrastructure: Triggers, Observability, and Enterprise Readiness

Operating agentic workflows at scale requires robust operational mechanics that extend beyond passive request-response cycles. AgentKit incorporates event-driven triggers, enabling agents to spin up immediately when real-world events transpire—such as receiving an email, detecting a webhook payload, or monitoring a database change—eliminating the need for custom polling infrastructure and background workers.

For enterprise risk and compliance officers, visibility is non-negotiable. AgentKit logs every tool execution, capturing comprehensive telemetry on who triggered the action, the input arguments passed, the API response received, and the exact timestamp. These audit streams can be exported directly to enterprise Security Information and Event Management (SIEM) systems. Paired with Bring Your Own Key (BYOK) support through cloud key management systems like Google Cloud KMS, private and self-hosted deployment models, and SOC 2 alignment, AgentKit establishes the governance backbone needed for high-compliance industries.

Industry Impact

The launch of AgentKit underscores a broader maturation across the generative AI landscape: the transition from raw model capabilities to operational infrastructure. Anthropic's introduction and open standardization of the Model Context Protocol (MCP) signaled a shift toward standardized agent-to-tool communication. However, protocol standards alone do not handle the operational complexities of multi-tenant identity, credential rotation, rate limiting, and regulatory compliance.

Scalekit's introduction of a specialized MCP gateway establishes a new category within the AI infrastructure stack—Agent Identity and Access Management (Agent IAM). By decoupling authentication and tool integration from core model logic, organizations can build agentic applications significantly faster without building bespoke integration layers for every SaaS tool. As autonomous agents increasingly take real-world actions with downstream financial and operational consequences, gateways providing strict identity delegation and complete observability will become an indispensable component of modern software development.

Frequently Asked Questions

What is AgentKit by Scalekit and how does it function?

AgentKit is an MCP gateway built by Scalekit to manage tool calling, authentication, and permissions for production AI agents. It provides a managed proxy between AI agent frameworks and external software, allowing agents to interact with over 600 applications and 30,000 tools while ensuring all calls are securely authenticated, logged, and isolated.

How does AgentKit protect user credentials during tool calling?

AgentKit isolates credentials within a dedicated, encrypted vault, completely removing access tokens from the agent's code, inference prompts, and application logs. When an agent triggers an action, AgentKit injects the authenticated user's scoped OAuth token at runtime, ensuring the agent acts strictly within that user's explicit permissions.

How do Virtual MCP Servers prevent tool bloat and context degradation?

Virtual MCP Servers allow developers to create custom groupings of tools tailored to specific agent tasks rather than loading an entire catalog into the LLM context. By restricting the tool definitions sent to the model, Virtual MCP Servers preserve context window space, decrease latency, lower token expenses, and minimize the chance of incorrect tool selection.

Related News

Gemini API Tool Retrieval Preview Gives Agents Access to Hundreds of Tools With Fewer Tokens
Product Launch

Gemini API Tool Retrieval Preview Gives Agents Access to Hundreds of Tools With Fewer Tokens

Google AI Blog announced that Tool Retrieval is now in preview in the Gemini API. The feature is presented as a way to give agents access to hundreds of tools while using fewer tokens. The announcement pairs those two points: the breadth of tool access and lower token use. It does not explain how Tool Retrieval works, specify how token use was measured, or quantify the reduction. It also does not provide an exact tool count beyond describing access to hundreds of tools. The stated release status is preview, and the announcement does not give a timeline for general availability.

LangSmith Tool Gateway Enters Public Beta for Secure, Governed AI Agent Tool Use
Product Launch

LangSmith Tool Gateway Enters Public Beta for Secure, Governed AI Agent Tool Use

LangChain announced that LangSmith Tool Gateway is in Public Beta. The gateway handles credentials, identity, and access for every agent tool call, with policies, approvals, and traces. The announcement frames secure, governed tool use around the individual call: its focus is on how agents access tools, rather than on what those tools or agents can do. The update identifies what the gateway covers but does not explain how credentials are managed, how access decisions are made, when approvals apply, or what information traces contain. The Public Beta is the only release status stated in the announcement.

Google’s Gemini Decision 1 Picks Answers Instead of Writing Them Across Five Input Types
Product Launch

Google’s Gemini Decision 1 Picks Answers Instead of Writing Them Across Five Input Types

Google’s Gemini Decision 1 is a model designed to pick answers rather than write them, according to Shittu Olumide’s report for KDnuggets. It reads text, images, video, audio, and PDFs in a single forward pass and supports up to 262,144 input tokens per request. The central distinction is between what the model reads and what it does with that input: it accepts several kinds of material, but its stated task is answer selection, not answer writing. The token figure describes the maximum input per request, not how much the model outputs. The report does not explain how answers are presented for selection or provide results showing how well the model performs.