A lot has been written about Model Context Protocol (MCP) lately. Rather than changing what AI can do, MCP changes how AI interacts with other systems. MCP isn't giving AI new capabilities. It's giving it a better way to ask for help.
In our demo above, Claude connects to DeepJudge to respond to an RFP. Claude reads the RFP, recognizes it needs relevant prior work from the firm's document history, reaches out to DeepJudge via MCP, and gets back exactly what it needs to draft a response. There’s no need to copy-paste, switch tabs, or manually retrieve information.
That's MCP in practice. It's an open standard, introduced by Anthropic, that allows an LLM like Claude to connect to an external application. The application (DeepJudge in our example) responds, the model understands, and they work together without a developer writing a single line of integration code.
MCP is genuinely useful. But it’s not magic, and it's worth being precise about why.
What didn't change in that demo: Claude isn't smarter. DeepJudge isn't doing anything new. The underlying capabilities are identical. MCP just made it frictionless for them to work together.
Where MCP differs from a traditional API is the audience. Traditional APIs are built for developers, who read the docs, write the code, and map the endpoints. MCP is built for LLMs. The model does the discovery itself. This lowers the cost and complexity of executing workflows that were previously possible, but difficult to operationalize.
There is a trade-off, though—MCP is narrowly scoped. It’s not a general-purpose integration layer, and it’s not a replacement for traditional APIs. Think of a traditional API as the infrastructure that keeps everything running, and MCP as the layer that lets an AI system understand what tools are available and use them appropriately when needed.
The "Schema Negotiation"
In a traditional API, a developer has to read documentation and hard-code the connection. With MCP, the LLM says, "Hey, what can you do?" and the server responds, "I'm an enterprise search engine; I can find internal documents." The AI then understands how to use those tools on its own, without a single line of new code from the user.
Is MCP Replacing APIs?
Short answer: No. MCP is quite restrictive in how it handles data. Traditional APIs are still the "heavy lifters" of the internet, handling large data transfers and complex system coordination.
Think of it this way:
For legal workflows, the implication is clear. AI systems only add value when they can reliably access relevant prior work with sufficient context. MCP makes that access easier, but it doesn’t change what the systems themselves can do.
Explore how firm-wide retrieval supports real legal workflows with MCP—without moving data or reworking governance.