MCP Server Quickstart

Overview

The Stack Internal MCP server connects an MCP-compatible assistant to indexed Stack Internal knowledge. It is a read-only server: the current implementation supports searching for nodes and retrieving a node by ID. It does not provide tools to create, edit, publish, vote on, or delete content.

Before you start

You need:

• An MCP-compatible client, such as Claude Code, Codex, GitHub Copilot, Gemini CLI, or a client that supports Streamable HTTP.
• A Stack Internal account for your workspace.
• Access to the MCP server. If the connection is unavailable, ask your Stack Internal administrator to confirm that MCP access is enabled for your organization.

Connect to the server

1. Add the server URL to your MCP client: https://mcp.stackinternal.com (https://mcp.eu.stackinternal.com for users in the EU).
2. Choose Streamable HTTP when the client asks for a transport type.
3. Complete the sign-in flow in your browser when your client opens it.
4. Start a new conversation and ask a question about knowledge indexed in Stack Internal.

Use the current production URL exactly as shown. The service is hosted at the root path, so do not add /mcp or an API version to the URL.

For detailed configuration instructions, read the MCP Server Overview article.

Use Stack Internal knowledge

After you connect, your assistant can use two read-only tools.

search_nodes
Searches indexed knowledge nodes. Provide a clear search phrase. You can ask for fewer or more results, up to 100, and choose lexical or semantic search when your client exposes those options.

Example requests:

• "Search Stack Internal for the incident-response process."
• "Find up to 10 results about employee onboarding using semantic search."
• "Search for the current guidance on customer data retention."

Search results can include the node ID, title, source, source ID, score, content ID, and source URL when those fields are available.

get_node
Retrieves one indexed node by its node ID. Use the ID from a search result when you need the full available content and metadata.

Example request:

• "Open the node with ID 12345 and summarize the key actions."

A returned node can include its source, content ID, URL, classification signals, dates, author, trust information, relationships, and text when that information is available.

Work effectively with the server

Start broad, then narrow your question. For example, search for “incident response” first. Then ask your assistant to open the most relevant result and cite or summarize it.

Use natural language that names the topic, system, document type, or time period you need. For example:

• “Find the newest operating procedure for customer escalations.”
• “Search for the design decision about tenant isolation.”
• “Show the source link and summarize the node about release approvals.”

Treat results as source material. Open the most relevant nodes and verify important decisions against the linked source before acting on them.

Troubleshooting

If a search doesn't find what you need:

• Try shorter, more specific terms.
• Search for a system, team, or document name.
• Ask for semantic search if your client lets you choose a search algorithm.
• Check whether the source has been ingested into Stack Internal.

If you can't connect:

• Confirm that the server URL is https://mcp.stackinternal.com (https://mcp.eu.stackinternal.com for users in the EU).
• Confirm that your client supports Streamable HTTP.
• Complete the browser sign-in flow, then retry the connection.
• Ask your administrator whether MCP access is enabled for your organization.

If you have questions or issues, reach out to our support team for help.