Licensing and monetization
MCP Access vs AI Data Licensing: Two Different Permissions
By SourceX Editorial · Updated
Short answer
Clarifying runtime access, export and training/evaluation permissions as separating scope decisions. Connecting a tool is about live access for your own team; letting an AI company use your records is a separate, company-approved decision.
- 01Your work records
- 02Personal details removed
- 03You approve the release
- 04Licensed to AI developers
What this means for your business#
This guide is written for CEO or founder, COO or CFO or CTO for technical scope. It covers clarifying runtime access, export and training/evaluation permissions as separating scope decisions — and what that means for a business deciding what its AI tools may reach.
Worth checking before you connect#
- Treat a connection and a license as two separate approvals, always
- Never let live access stand in for a signed data agreement
- Inventory systems and years of history before any licensing talk
- Have counsel review any draft agreement before signing
- Expect buyers to ask about rights and privacy work, not connections
Access is not licensing#
SourceX handles the licensing side: rights review, preparation, your approval, then delivery to the AI buyer under a signed license. Your MCP connections stay your own — we never ask for them.
Two different permissions, side by side#
| Live access (MCP) | Prepared data release (license) | |
|---|---|---|
| What it reaches | Current records in the connected tool, within one user's permissions | A complete historical copy, prepared and reviewed |
| Who approves | The account owner approves each connection | The company approves exactly what is released, once per deal |
| What it's good for | Everyday questions, lookups and drafting | Training, evaluation and analysis by an AI company, on agreed terms |
| How long it lasts | Reads happen only while connected | Runs for the signed license term |
Check your fit
Q1 / 05 · COMPANY SIZE
How many full-time employees at your peak?
What is MCP, the Model Context Protocol?#
MCP (Model Context Protocol) is an open protocol that gives AI applications a standard way to connect to external data sources and tools. A system exposed once as an MCP server can be used by AI applications that support the protocol.
The specification defines three participants that exchange JSON-RPC 2.0 messages: hosts, clients and servers. Its authors say MCP takes inspiration from the Language Server Protocol, which standardized how code editors add support for programming languages. Messages travel over one of two standard transports: stdio, where the client launches the server as a local subprocess, or Streamable HTTP, where each message is an HTTP POST to a single endpoint.
| Participant | Role in the specification | Business example |
|---|---|---|
| Host | The LLM application that initiates connections | A desktop AI assistant or an AI-enabled code editor your staff use |
| Client | The connector inside the host that talks to one server | The assistant's link to your ticketing system |
| Server | The service that provides context and capabilities | A server that exposes your Jira or Zendesk records to the assistant |
| Transport | How messages are carried between client and server | A local subprocess (stdio) or a remote HTTP endpoint (Streamable HTTP) |
What can an MCP server expose, and who controls each part?#
An MCP server can offer three kinds of features, and the specification assigns each a different controller.
Because tools are model-controlled, the specification says there should always be a human in the loop with the ability to deny tool invocations.
| Feature | Controlled by | What it is | Example from the specification |
|---|---|---|---|
| Prompts | The user | Templates the user chooses to invoke | Slash commands, menu options |
| Resources | The application | Context and data the client attaches and manages | File contents, git history |
| Tools | The model | Functions the model can call to retrieve information or take actions | API POST requests, file writing |
What does the MCP specification say about consent and access to your data?#
The specification asks implementers for explicit user consent before data is exposed or tools are run, but it also says MCP cannot enforce those principles at the protocol level. What actually protects your records is how the host, the server and your own account permissions are built and configured.
- Hosts must obtain explicit user consent before exposing user data to servers, and must not transmit resource data elsewhere without user consent.
- Hosts must obtain explicit user consent before invoking any tool, and tool descriptions should be treated as untrusted unless they come from a trusted server.
- Servers should not be able to read the whole conversation or see into other servers; the full conversation history stays with the host.
- Authorization is optional in MCP. When it is supported over HTTP it follows OAuth 2.1, and a server must accept only access tokens issued for it; local stdio servers instead take credentials from their environment.
- None of these rules grants a right to use your records for training or evaluation; that is a separate, written decision.
Sources for this section
- The MCP specification defines MCP as an open protocol connecting LLM applications to external data and tools through hosts, clients and servers exchanging JSON-RPC 2.0 messages, and sets consent principles for data access and tool use. Source
- The MCP specification defines two standard transports: stdio for client-launched local servers and Streamable HTTP for remote endpoints. Source
- The MCP specification assigns prompts to user control, resources to application control and tools to model control. Source
- The MCP tools specification says a human should always be able to deny tool invocations. Source
- The MCP architecture says servers should not read the whole conversation or see into other servers. Source
- The MCP authorization specification makes authorization optional, bases HTTP authorization on OAuth 2.1, and requires servers to accept only tokens issued for them. Source
Frequently asked questions
Does connecting an MCP server let an AI company train on our records?
No. MCP gives the assistant you connected a way to read what its user can already read, during your own sessions. Use of your records by an outside AI company is a separate decision that happens through a signed agreement — never through a connection.
What should we do first if we are considering MCP data licensing?
Write down which systems hold the records you care about, who owns the approval, and what is read-only today. That short inventory answers most of the questions that come up later.
Can SourceX help us assess our records for AI companies?
Yes. SourceX reviews your rights, prepares a clean historical copy and shows you exactly what would leave your company for approval before anything is shared. SourceX works with companies that had 50 or more full-time employees at their peak (contractors excluded): U.S. companies first, plus companies in Canada and the EU, and in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 90% of the data is in English.
Sources
Vendor and ecosystem pages used while writing this guide. For exactly what any vendor's MCP server exposes, their own documentation is the source of truth.
Related
General information, not legal advice. Editorial policy.
Tell us what your models need
Share scope, volume, language, format, timing and licensing requirements.