Plugging LeadLex into Claude: Inside the Anthropic MCP Integration

Most enterprise AI assistants run into the same wall the moment they touch real work. The model is fluent. The conversation is plausible. But the answers are generic, because the assistant has no access to the firm's actual data — its clients, matters, contacts, mandates and history. Without that grounding, "AI for the firm" reduces to a smarter search engine on the public internet.

The Model Context Protocol — MCP — is Anthropic's attempt to fix that wall. Released as an open standard in late 2024, MCP defines a common way for AI assistants like Claude to securely connect to the systems where work actually lives: CRMs, document repositories, calendars, ticketing systems, code. It is to AI agents what USB was to peripherals: a single interface that any client can speak.

LeadLex is now exposed as an MCP server.

What an MCP-aware Claude can do for a partner

In practice, any MCP-aware client — Claude in the desktop app, Claude inside Microsoft Teams, Claude Code, or any future agent that adopts the standard — can read from and act on the firm's LeadLex workspace, under the same permissions a human user would have.

A partner using Claude can ask, in their own words:

• "What do I need to know before my call with the GC at [Client]?"
• "Find the three closest analogous matters we have done for renewable energy clients in Germany, and pull the relationship maps."
• "Draft a follow-up to the conversation I had yesterday with [Contact] and log it to the matter."

Each of those queries hits real firm data: clients, contacts, prior mandates, conflict records, calendars. Each one returns an answer that is specific to the firm, the client and the partner — not the public internet's best guess about what a generic firm might do.

The reason this is interesting is not the integration. Integrations have always been buildable. It is the architectural shift underneath.

The architectural shift

In the old model, the CRM was a destination. Lawyers visited it, looked things up, and updated fields. In the AI-native model, the CRM is an exposed surface. Agents reach into it on demand, read what they need, write what they have learned, and return to the channel where the lawyer is working — Teams, Slack, WhatsApp, the desktop. The CRM becomes infrastructure rather than a tool.

This has three consequences for firms thinking about their stack.

First, there is no longer a meaningful trade-off between "centralised data" and "where the work happens". Centralised data lives in LeadLex. Access to it lives wherever the partner is. The data layer and the experience layer have been decoupled.

Second, the choice of AI client becomes less load-bearing. A firm that standardises on Claude today is not locked in. Any future agent that speaks MCP will be able to work with the same firm context, because the data is exposed in a standard way. The integration is into the standard, not into the assistant.

Third, security and permissioning move with the data, not the assistant. MCP carries the user's identity and scope. A junior associate's Claude session sees what they are entitled to see — no more, no less. This is the bar enterprise legal IT has rightly insisted on, and it is now achievable without bespoke per-tool engineering.

Why this fits how LeadLex was built

LeadLex was built with this future in mind. The product has always assumed that the partner's primary interface to firm data would be a conversation — in chat, in email, in their own voice — rather than a form. MCP is the technical layer that makes that assumption coherent across every AI client a firm might adopt.

For partners, the experience is simple: ask the question; get an answer grounded in the firm. For IT, it is even simpler: one data layer, one permissioning model, every AI assistant in the room.

That is what AI-native infrastructure looks like.

Related: Why the CRM should live inside WhatsApp, email and Teams. What's on the integrations roadmap next.