Delinea unveils open-source MCP Server to secure AI agent access

Delinea has launched its open-source Model Context Protocol (MCP) Server, enabling developers to embed secure, policy-driven access to credentials and workflows into their custom AI agents.

The MCP Server package, now freely available on GitHub, is designed to bridge the gap between AI agents and practical real-world use, connecting AI models with tools, data, and processes on the Delinea Platform. As organisations increasingly deploy AI agents to drive software development and IT operations, managing secure connections to external tools and sensitive data has emerged as a complex challenge.

Traditionally, integrating AI with enterprise systems has required either building custom connectors for each application or embedding credentials directly within large language model (LLM) prompts, approaches that increase the risk of unauthorised access and create compliance headaches. Delinea aims to address this with MCP Server, which governs every access request with industry-standard identity context, privilege, and policy checks.

Security concerns

As AI adoption accelerates within organisations, there are growing concerns around granting appropriate access and maintaining regulatory compliance. Delinea has emphasised the importance of containing the risks associated with AI agents, such as unauthorised access or erroneous requests.

"AI agents are revolutionising innovation, but they also introduce security and compliance challenges," said Phil Calvin, Chief Product Officer at Delinea. "To mitigate risks like access request hallucinations and vibe hacking, it's crucial to enforce identity context at every interaction. MCP Server ensures AI agents operate within secure boundaries, using ephemeral tokens and policy checks to maintain traceability and protect credentials."

MCP Server governs access by employing ephemeral tokens, meaning credentials are temporary and used only for specific requests. This design minimises the risk of credential leakage and ensures that all actions are auditable. According to Delinea, this is especially important for companies operating under strict compliance frameworks, as it supports both transparency and security.

Productivity and compatibility

Organisations often spend considerable time and resources building and maintaining custom API connectors. MCP Server aims to streamline this process. By using open standards like OAuth and embedding security best practices, Delinea claims that organisations will see faster deployment times and reduced engineering overhead. The package is compatible with leading vendors and also offers experimental connectors for open-source AI projects such as Claude and ChatGPT, allowing customers to choose tools that fit their broader AI strategy.

The company highlighted four major benefits of MCP Server: reducing risk by ensuring AI never receives unrestricted access to sensitive data; improving productivity by removing the need for custom connectors; supporting a future-proof AI strategy through open standards; and enabling practical AI adoption via direct, natural-language management of users, secrets, and access policies.

Flexible use cases and future direction

Delinea states that the flexibility offered by the open-source package will help organisations adapt as AI ecosystems continue to evolve. MCP Server allows developers to create AI agents that can manage users, groups, secrets, and access requests using natural language or advanced interfaces, including voice.

The MCP Server is described as Delinea's first official open-source package aimed at extending the range of applications for its platform. Its compatibility with popular open standards and support for a variety of AI agent vendors is designed to offer customers freedom as technology choices develop over time.

MCP Server is currently available for no cost on GitHub and can be integrated into existing AI systems. The company is positioning this release as an entry-point for organisations looking to manage AI agent access and compliance requirements in their digital environments.