How to Create an MCP Server From Multiple Sources

To create an MCP server from multiple sources, add each source to the same server, for example a documentation site and the GitHub repository behind it. In Context MCP Studio you add both in the wizard and deploy once. Your AI client still connects to one URL, and each answer cites the source it came from. Add a second source only when it answers a question the first cannot.

By the end of this guide you will have one server over two sources, a question that should cite each one, and a way to adjust things when the wrong source keeps answering.

What a multi-source server is

It is one hosted endpoint over more than one body of material. The client does not need a second connection, and you do not manage two URLs. What you do need is a reason for the second source. If you cannot write a question only that source can answer, it is not ready to add.

This guide follows one example throughout. A payments product publishes its maximum upload size in the docs, while the sample client that handles a rejected upload lives in the repository. Each source answers one of those questions and has nothing to say about the other, which is exactly when two sources belong on one server.

Give each source a job

Documentation answers the contract. Methods, required fields, limits, and the sentences a customer is allowed to rely on all live here. "What is the maximum upload size?" should cite the docs.

A repository answers the implementation. The function that checks the file, the sample that handles the error, and the comments beside them live here. "How does the sample handle a rejected file?" should cite the repository.

A PDF answers the rule that was never published as a site. A handbook or a spec export fits this role. How that upload works is covered in Turn a PDF into an MCP server.

Another product's MCP endpoint answers from its own docs. GitBook, for example, publishes one alongside each site, and you can put it beside your repository on one server, as shown in Your GitBook docs and your code behind one MCP endpoint. That connection is used for retrieval only; it does not expose the remote server's action tools to your client.

Decide who can reach the server

A private GitHub repository makes the whole server private. The server stays private until that source is disconnected, because it can return content a stranger could not fetch. Adding sources covers the GitHub sign-in, and Who can reach your MCP server covers the access token that clients use afterward.

How to build the server in MCP Studio

Open the wizard with your two questions already written down. You do not write any code.

1. Name it for the job, not the inventory. Payments is clearer than Docs plus repo.
2. Add the documentation URL first, then the repository. Each source should match one of the questions you wrote down.
3. Choose tools that match the questions. API reference lookup suits the docs, code example lookup suits the repository, and search covers both. The tool guide explains how they differ.
4. Deploy and wait until both sources finish indexing. An answer from a source that is still indexing is not a fair test.

Copy the URL from the setup page:

https://appatools.com/mcp-studio/api/mcp/your-server-name

The free tier includes two sources on each server. Each source you add beyond those two is a one-time $3 add, described in the billing guide.

How to check that it works

Ask the two questions separately, each in a fresh chat with the server turned on.

1. Ask "What is the maximum upload size?" and expect a citation from the docs.
2. Ask "How does the sample handle a rejected file?" and expect a citation from the repository.

Open each citation. The page should be the one you had in mind, not a neighboring README that mentions uploads in passing. Then ask something neither source covers.

You: What is our incident escalation policy?

Assistant: Nothing in this server's content covers incident escalation. The closest material is the upload guide, which is a different subject. You would need to add that document as a source.

Two correct citations and one clear miss tell you both sources are earning their place.

If the wrong source keeps answering

Prefer a source with a retrieval rule. When the repository keeps answering the policy question, write a rule that prefers the docs for limits and authentication. A rule can prefer a source or require that a passage from it is included, but it cannot invent an answer the sources do not contain. The retrieval rules guide shows how to write one.

Split the server when you cannot tell which source was wrong. Unrelated sites on one endpoint make the test impossible, so group sources by job instead. That is the same advice as Ship a docs MCP built around one real task.

Frequently asked questions

How many sources can one MCP server have?

On the MCP Studio free tier, each server includes two sources. You can add more sources to a server for a one-time $3 each, and a source you have paid for stays on the server.

Can I combine documentation and a GitHub repository in one MCP server?

Yes. This is the most common multi-source setup: the documentation answers questions about limits and required fields, and the repository answers questions about how the code implements them. Your AI client connects to one URL and each answer cites whichever source it came from.

Should I use one MCP server or several?

Use one server per job. Sources that answer the same task from different angles belong together, while unrelated sites are better on separate servers. A focused server is easier to test, because you can tell which source produced a wrong answer.

How do I know which source answered a question?

Every passage the server returns links back to the page it came from. Open the citation in your AI client's reply to see whether it points at the docs or at a file in the repository.

What happens when I remove a source from a server?

Disconnecting a source deletes the content indexed from it, including excerpts kept for analytics. The other sources on the server stay in place and keep answering.

Build a multi-source server with MCP Studio

Open MCP Studio with your two questions written, add the docs and the repository, and deploy. It takes no code, and asking each question once shows from the citations whether the second source earned its place. Start building.

Where to go next

• Give an AI agent access to your GitHub repository
• Turn your documentation into AI context
• Adding sources
• Retrieval rules