7 - MCP | Dosu

The Dosu MCP (Model Context Protocol) server lets AI coding assistants search your organization's knowledge base directly. Connect tools like Cursor, Claude Code, or VS Code to get answers from your documentation, code, and conversations without leaving your editor.

Getting Started#

When you create a Dosu organization, a default MCP deployment is automatically created for you. This deployment is named "Default MCP" and is linked to your organization's data sources.

To find your MCP deployment:

Go to Settings → MCP
You'll see your default deployment listed with its deployment ID and connection instructions

Creating Additional MCP Deployments#

If you need additional MCP deployments with different data source configurations:

Go to Settings → MCP
Click Add New
Enter a name for your MCP server (e.g., "Engineering Team MCP")
Select which data sources the MCP server should have access to
Click Create

After creation, you'll see your deployment ID and connection instructions for each supported client.

Connecting Your AI Assistant#

Claude Code#

Claude Code supports OAuth authentication natively. You can connect using either a path-based URL or headers.

Option 1: Path-based (simpler)

claude mcp add --transport http dosu https://api.dosu.dev/v1/mcp/deployments/<your-deployment-id>

Option 2: Header-based

claude mcp add --transport http dosu https://api.dosu.dev/v1/mcp \
  --header "X-Deployment-ID: <your-deployment-id>"

Replace <your-deployment-id> with your deployment ID. Claude Code will prompt you to authenticate through your browser—no API key required.

Cursor#

Cursor currently requires an API key for authentication. You can connect using either a path-based URL or headers.

Option 1: Path-based (simpler)

Add the following to your Cursor MCP settings file (.cursor/mcp.json in your project root or global config):

{
  "mcpServers": {
    "dosu": {
      "url": "https://api.dosu.dev/v1/mcp/deployments/<your-deployment-id>",
      "headers": {
        "X-Dosu-API-Key": "<your-api-key>"
      }
    }
  }
}

Option 2: Header-based

{
  "mcpServers": {
    "dosu": {
      "url": "https://api.dosu.dev/v1/mcp",
      "headers": {
        "X-Deployment-ID": "<your-deployment-id>",
        "X-Dosu-API-Key": "<your-api-key>"
      }
    }
  }
}

Replace <your-deployment-id> and <your-api-key> with your values. You can create an API key in Settings → API Keys.

Tip: You can also click the "Configure on Cursor" button in your MCP deployment settings to automatically open Cursor with the configuration pre-filled.

VS Code and Other Clients#

MCP support in VS Code depends on your AI extension. You can specify your deployment ID in the URL path or via headers.

Option 1: Path-based endpoint (simpler)

Transport: HTTP

Endpoint: https://api.dosu.dev/v1/mcp/deployments/<your-deployment-id>

Authentication: OAuth 2.0 (recommended) or API Key

Required Headers:

X-Dosu-API-Key: <your-api-key> (only if OAuth is not supported)

Option 2: Header-based endpoint

Transport: HTTP

Endpoint: https://api.dosu.dev/v1/mcp

Authentication: OAuth 2.0 (recommended) or API Key

Required Headers:

X-Deployment-ID: <your-deployment-id>
X-Dosu-API-Key: <your-api-key> (only if OAuth is not supported)

Refer to your extension's documentation for adding MCP servers.

Authentication#

Dosu MCP supports two authentication methods:

OAuth 2.0 (Recommended)#

For clients that support OAuth, Dosu uses OAuth 2.0 with Dynamic Client Registration. The client will prompt you to authenticate through your browser. This is the recommended approach—it requires less setup and automatically rotates credentials.

API Key#

If your client doesn't support OAuth, you can fall back to API key authentication. Create an API key in Settings → API Keys and include it in the X-Dosu-API-Key header.

Available Tools#

When connected to Dosu's MCP server, AI assistants can use these tools:

Choosing the Right Search Tool#

init_knowledge – Your starting point for any task. Searches curated knowledge (approved answers and topics). Use this first to get comprehensive context.
search_documentation – Searches raw documentation sources like Notion, Confluence, and GitHub wikis. Use this when init_knowledge didn't find what you need or you need documentation specifically.
search_threads – Searches discussions and conversations including GitHub issues, Slack threads, and other discussion sources. Use this when looking for bug reports, past decisions, troubleshooting history, or community discussions.

If no results are found, try broadening your query, using different keywords, or searching with a different tool.

All Tools#

init_knowledge – Search your knowledge base for task-relevant information. Use this at the start of any task to get context from curated knowledge (approved answers and topics) rather than raw documentation or discussion threads.
search_documentation – Search across your documentation sources including Notion, Confluence, GitHub wikis, and knowledge base articles. Use this to find how-to guides, API references, and architectural decisions. Be specific with your queries and include key terms or feature names.
search_threads – Search through GitHub issues, Slack conversations, and other threaded discussions. Use this to find past solutions, bug reports, error messages, and team decisions. Useful for troubleshooting history and community discussions.
fetch_source – Retrieve the full content of a source by its source_id (provided in search results). Use this after searching to get complete details when the summary isn't enough. You can specify line ranges to fetch specific sections of large documents.
save_topic – Save an important topic to the knowledge base for future reference. A topic represents a durable, high-level engineering or product theme that emerged from your research. This should be used for a single feature, theme, or component that exists in the current codebase. Topics are surfaced to other users in your organization and can be reviewed by maintainers.
greet – Test the MCP connection and verify everything is working. Use this only for debugging connection issues.

All search tools return the most relevant results based on your query, along with links to the source material.

Managing MCP Deployments#

To view, edit, or delete your MCP deployments, go to Settings → MCP.

From here you can:

View connection details and deployment IDs
Update which data sources are connected
Rename deployments
Delete deployments you no longer need

Troubleshooting#

"No data sources found for this deployment"#

Make sure you've connected at least one data source to your MCP deployment. Go to Settings → MCP → [Your Deployment] → Manage Connections.

"Invalid or expired API key"#

Create a new API key in Settings → API Keys. Make sure you're using the full key (starting with dosu_) in your configuration.

"Source not found or access denied"#

The source may have been removed from your connected data sources, or the search results have expired. Try searching again.

Connection Issues#

Verify your deployment ID is correct
If using API keys, check that your key hasn't been revoked
Try the greet tool to test basic connectivity