Agentic AI
Docusign's MCP Server moves to general availability after learning how AI agents access enterprise software safely
Moving Docusign's contract tools from beta to production revealed complex challenges around agent reliability, permission enforcement, and multi-client support that went far beyond simple API integration.
1/3
The New Stack
Integrating an AI agent with enterprise software might seem straightforward on paper. In practice, ensuring that agents can reliably find the right information, execute tasks correctly, and respect user permissions requires far more nuance than initial proof-of-concepts suggest.
Docusign's experience bringing its Model Context Protocol (MCP) Server from open beta in May 2026 to general availability on September 30, 2026 illustrates the gap between demos and production-ready AI integrations. The platform lets AI assistants from vendors like Anthropic, OpenAI, Google, and Microsoft work with Docusign's contract capabilities—retrieving documents, checking status, or preparing agreements for signature—without forcing users to leave their preferred AI environment.
From Traditional APIs to Agentic Workflows#
Traditional REST APIs require developers to know in advance which specific requests they'll send. An AI agent operates under a fundamentally different model: users describe goals in natural language, and the agent decides which tools to invoke and in what sequence.
MCP, an open standard, bridges this gap by exposing a server's capabilities in a way that AI applications can discover and dynamically invoke. Docusign's MCP Server makes selected contract management features available within the AI tools its customers already use, extending its Intelligent Agreement Management (IAM) platform into these new environments.
What Beta Testing Exposed#
Shipping software to a small cohort of beta users who anticipate likely scenarios differs markedly from releasing to a wider audience with unpredictable usage patterns. Docusign's CTO Sagnik Nandy observed that "there's a big difference between building tools and solving actual customer problems." As adoption broadened, the company encountered complexities that early testing hadn't fully surfaced.
Users wanted to upload and work with documents stored outside Docusign's systems. They requested account-level administrative controls to enable or disable certain MCP capabilities. They posed prompts in ways the development team hadn't fully anticipated, and different AI clients handled the same request differently. These findings pushed Docusign to expand supported workflows, add local file upload capability through the MCP Apps interface, and introduce account-level configuration options.
The company also had to harden reliability. Since changes to underlying AI models or client behavior can affect tool interaction even when the integration itself remains unchanged, Docusign invested in guided interfaces, automated evaluations, and security testing to validate how agents select and use its tools.

