What We Learned from Evaluating MCP-Based Workflows for Authoritative Legal AI

At LexisNexis, we believe legal AI must do more than generate plausible answers. It must produce work lawyers can verify, defend, and trust.

That principle – AI held to the standard of law – guides every technology decision we make, including which external integrations help us better meet our customers’ need for legal authority, and which require further development before they are ready for customer-facing legal workflows.

Model Context Protocol, or MCP, is an open standard that enables AI systems to connect with complementary tools, data sources, and resources to help users complete tasks. As MCP-based integrations become an increasingly common architecture across legal AI, we have been testing how they perform in practice – especially in workflows where accuracy, attribution, consistency, and defensibility matter.

Our Approach: Continuous Evaluation, Selective Integration

Lexis+ with Protégé is an ecosystem-based legal AI platform. It brings together capabilities developed by LexisNexis, as well as capabilities contributed by technology partners, including Anthropic, where they demonstrably serve our customers’ needs.

The launch of the next evolution of Lexis+ with Protégé and the announcement about the integration of Anthropic’s Claude legal plugin suite into the platform both reflect that philosophy. We believe in ecosystem-based innovation, and we integrate external capabilities selectively – when they strengthen the legal work our customers need to perform.

That also means we test rigorously before we ship. We work closely with partners across the AI ecosystem to evaluate emerging capabilities against the standards legal work demands: citation accuracy, output consistency, grounding in authoritative sources, and clear attribution between what LexisNexis content supports and what a model generates independently.

MCP Testing: Our Criteria for Integration

When evaluating potential MCP connector integrations, we run methodical side-by-side comparisons across three configurations: the LLMs alone, LLMs with our MCP connector, and Lexis+ with Protégé.

In testing, if an LLM has access to our MCP connector – and introduces additional legal interpretations, adjacent statutes, or citations not provided by LexisNexis – our criteria for moving forward with an integration is not met. The same is true if the resulting output could reasonably lead a user to believe the full response was grounded in LexisNexis authority, when portions were generated independently by a model.

In addition to attribution ambiguity, consistency is another important consideration. Asking the same legal question multiple times through an LLM should not produce meaningfully different outputs. We require the same performance of Lexis+ with Protégé; each response to a given prompt should remain consistent and tightly tied to authoritative support.

For legal work, this difference matters. It can be the difference between work product a lawyer can confidently validate and output that requires additional scrutiny before use.

Why Authoritative Grounding Is Non-Negotiable

We are careful not to present authoritative LexisNexis content alongside model-generated material in ways that could blur attribution or undermine user confidence. This is not a competitive posture. It is a way of ensuring that we assist our customers in meeting their professional obligations.

Lawyers rely on LexisNexis because our legal AI is verifiable, citable, and connected to trusted legal authority. When a user cannot clearly distinguish between what LexisNexis has verified and what a model has generated independently, the trust that makes our legal AI useful is weakened – and the lawyer’s work product may carry risk the user cannot easily see.

AI without deep legal authority and built-in ability to verify can accelerate work. It can also accelerate the need for review. The pace of AI development is genuinely exciting, and we are deeply engaged with it. Our customers’ professional reputations depend on the verifiability, attribution, and integrity of every AI response we deliver. That responsibility remains constant no matter how quickly the technology moves.

Where We Go from Here

Continuous Lexis+ with Protégé legal AI testing is underway with all our important and valued technology partners. We embrace this cutting-edge work with leaders across the AI ecosystem to meet customers where they are, while ensuring that their need for authoritative legal work remains fully supported and uncompromised.

Our objective is to ensure that innovation in legal AI empowers lawyers to achieve better outcomes – in a way they can fundamentally trust and directly verify.