The Connected M&A Technology Stack: Using MCP to Link Due Diligence, Legal Research and Drafting

Quick answer: The Model Context Protocol (MCP) is an open standard that lets AI assistants connect securely to other systems using the user's own permissions. In M&A, it means a lawyer's assistant can read structured findings from a due diligence workspace, check a point in a legal research service and help draft the SPA mark-up, without copying documents between tools. Each specialist system does what it does best, and the work stays connected.

Key takeaways

• No single tool does everything well across a transaction. The stack is the product.
• MCP has moved fast in legal: document management, legal research, data rooms and diligence tools now support it.
• Connection raises governance questions. Read-only access, user-level permissions and logging are the controls to insist on.

The fragmented deal

Follow a finding through a transaction today.

An associate spots a change of control clause in the data room. She records it in the review tool. She copies it into the report in Word. The partner reads the report and emails the SPA drafting team. Someone checks the legal position in a research database. Someone else adds a condition precedent to the SPA and a line to the closing checklist.

That is five systems, with the finding re-typed or pasted at each step and context lost every time. By the time it reaches the SPA, nobody can click back to the clause that started it.

Generative AI did not fix this on its own. Most firms now have several AI tools, and each one knows only what it has been given. Writing in Artificial Lawyer this year, a product manager at Legatics described two gaps holding legal AI back, both about connectivity: AI lacks context from the firm's systems, and it cannot act across them.

What MCP is, in plain terms

MCP is an open standard, originally developed by Anthropic in late 2024 and since donated to the Linux Foundation. It defines a common way for AI applications to exchange context with other systems.

Three terms are enough:

• MCP server. A system that exposes some of its data or actions in a standard way. For example, a due diligence workspace exposing deal findings.
• MCP client. An AI application that can connect to servers. For example, Claude, or a firm's own assistant.
• Tools and permissions. What the server lets the client do, and on whose behalf.

The comparison often made is with HTTP for the web: one standard interface in place of a custom integration for every pair of systems.

Why 2026 was the year it reached legal

The pace has been striking.

• Document management. iManage launched an MCP server in May 2026 offering permission-aware, logged access to its content. NetDocuments announced a collaboration with Anthropic so Claude can search and retrieve from a firm's own documents.
• Legal research. Thomson Reuters connected CoCounsel Legal to Claude through MCP in May 2026.
• Data rooms. Ideals added MCP connectivity so assistants can work with a data room without documents being exported.
• Deal data. Harvey and Datasite announced a partnership to bring live deal data into Harvey's workflows.
• Due diligence. Emma Legal launched its MCP connector, available in Claude's connector directory and in Saga, with ChatGPT support pending.

The direction is clear. Firms will evaluate legal technology on how well it connects, as well as on what it does alone.

The connected stack, stage by stage

Deal stage	System of record	What it contributes
Document exchange	Virtual data room	The documents, access control, Q&A
Legal due diligence	Due diligence workspace	Classification, gap analysis, checks, source-linked red flags, report
Legal analysis	Legal research service	Authority on the points of law the findings raise
Precedents and know-how	Document management system	The firm's own templates and prior deals
Drafting and negotiation	Drafting tools and Word	SPA, disclosure letter, ancillaries
Signing and closing	Transaction management	Checklists, signatures, closing sets
Across all stages	AI assistant (MCP client)	Reasoning and drafting, drawing on each system with the user's permissions

What it looks like in practice

A senior associate is preparing the first SPA mark-up. In her assistant, connected to the firm's approved systems, she asks:

"List the high-risk findings on change of control and consent in Project Atlas. For each, give the contract, the counterparty and the recommendation."

The assistant reads the findings from the due diligence workspace. It does not re-review the data room. It retrieves what the team has already found and verified, with links to the sources.

"Draft a conditions precedent schedule covering the three contracts recommended for consent before closing, using our firm's standard CP clause."

The assistant pulls the precedent from the document management system and drafts the schedule.

"Is a deemed-assignment clause likely to be triggered by an indirect change of control under the governing law of the Customer A contract?"

The assistant consults the connected research service and returns the position with citations for her to check.

Three systems, one conversation, and every statement traceable to where it came from. The associate remains the one who decides what goes in the draft.

Why specialists connected beat one tool for everything

It is tempting to hope one vendor will cover the whole lifecycle. Three reasons the connected model is more likely to win in transactions:

1. Depth. Diligence, research and drafting are different disciplines. A system built around request lists, playbooks and red flag reports handles diligence better than a general assistant, and the reverse is true for open-ended drafting.
2. Ownership of the record. Each stage has a natural system of record. The data room holds the documents. The diligence workspace holds the findings. Duplicating them elsewhere creates version risk.
3. Choice. An open standard lets a firm change one component without rebuilding the rest, and lets different practice groups use the assistant they prefer.

Governance: what to insist on

Connection without control is a risk. MCP itself leaves a lot to the implementer, so ask every vendor these questions.

• Whose permissions apply? The assistant should see only what the signed-in user can already see. No shared service accounts with broad access.
• Read or write? Start with read-only. Be deliberate about any connection that can change records.
• How does authentication work? Look for standard sign-in with explicit user consent, and admin approval of connectors.
• What is logged? Access through the connector should be logged like any other access.
• Where does the data go? Data read by the assistant is processed under the assistant's terms. Check its retention, training and residency position.
• What about instructions hidden in documents? Deal documents come from third parties. Assistants should treat document content as data, and connectors should limit what actions are possible.
• Ethical walls. Confirm that matter-level restrictions carry through the connector.

An approved-connector list, owned by IT and risk together, is the simplest control a firm can put in place.

How Emma Legal fits the stack

Emma is the system of record for the legal due diligence itself, and it connects on both sides.

Upstream, Emma syncs with leading data rooms, including Datasite, SS&C Intralinks, Ansarada, Ideals, HighQ and Virtual Vaults, and with SharePoint, Box, Dropbox and Google Drive. Documents flow in from the live room.

Downstream, the Emma MCP connector lets an approved assistant read the deals a user can already open in Emma, including request list status, checks, red flags and report summaries. The connector is read-only. It cannot change documents, checks, risks or access rights. Each person authenticates with their own Emma account, and IT administrators can review what the connector can access before approving it.

Emma is available in Claude's connector directory and in Saga, and can be added to other MCP clients that accept a remote server. ChatGPT availability is pending review.

In practice, the diligence stays structured, verified and source-linked in Emma, and the findings become usable wherever the lawyer is drafting.

Getting started

1. Map your deal workflow and name the system of record at each stage.
2. Ask each vendor about MCP support and its permission model.
3. Choose the assistant or assistants your firm will approve as MCP clients.
4. Start with one read-only connection on one deal team.
5. Write the governance: approved connectors, logging, client-data rules.
6. Expand as the team builds confidence.

An example of a defensible connected workflow

An associate identifies an assignment restriction during diligence. The reviewed finding and source clause remain in the due diligence workspace. An authorized assistant retrieves the approved summary via a permissioned MCP connection. A lawyer checks applicable law in the firm's research tool, drafts a tailored closing condition in the document system and links it back to the finding for final review.

MCP is an interoperability standard, not a guarantee of confidentiality or legal accuracy. Apply least-privilege access, confirm what a connected model receives, test revoked permissions and log material actions. A read-only connector can still transfer sensitive content for processing under the downstream provider's terms.

Emma's role is to preserve the transaction's diligence record: IRL, checks, review decisions, citations and report. Its connector can supply approved facts to the assistant used for drafting and research. Broader legal AI may remain the firm's primary work environment. The objective is an auditable handoff among capable tools, not an artificial claim that one product can do everything.

Frequently asked questions

What is MCP in legal tech?
An open standard that lets AI assistants connect to legal systems, such as document management, research and due diligence tools, through a common interface and with user-level permissions.

Is MCP secure enough for client data?
The protocol is a transport standard. Security depends on implementation: authentication, permissions, read-only scopes and logging. Evaluate each connector.

Does connecting an assistant mean our documents are copied into it?
The assistant retrieves what it needs for a request. That content is then processed under the assistant's terms, so use enterprise versions with appropriate retention and training commitments.

Do we need MCP if we have one firmwide AI tool?
Probably. Even a broad tool needs to reach your data room, document management and specialist systems.

Which assistants work with Emma?
Claude and Saga today, plus MCP clients that support remote servers. ChatGPT is pending.

Connect Emma to the assistant your team already uses. Book a demo at emma.legal.

Sources

• Artificial Lawyer, MCP: the standard that decides legal AI's future
• Thomson Reuters and Anthropic MCP integration, as reported
• iManage MCP server launch, as reported
• NetDocuments collaboration with Anthropic
• Ideals, virtual data room providers compared
• Harvey and Datasite partnership
• Emma MCP overview
• Emma: set up Claude