Connect your coding agents to your team's knowledge through the Dosu MCP Server. This page covers what the connection gives you, how to set it up, and every tool the server exposes so your agents can retrieve and enrich knowledge as they work. For the full reference, see Dosu MCP Server.

What connecting it gives you#

Coding agents write better code when they can see how your team actually works. The Dosu MCP Server gives agents like Claude Code, Cursor, and VS Code Copilot a way to query your Libraries directly from the editor, so they pull architectural decisions, conventions, and past answers instead of guessing.

The result is fewer wrong assumptions, less copying and pasting context into prompts, and more consistent output across everyone using agents. As you grow from one repo to many teams, the same knowledge backs every agent at once.

Availability and prerequisites#

• The MCP Server is available today and works with any MCP-compatible client. Many organizations use tools like Claude Code, Cursor, and VS Code Copilot.
• You need a Dosu organization with at least one Library so the server has knowledge to serve. Create or open one under Libraries.
• A default MCP server is created with your organization. You can view it and create more under MCP servers.
• API key authentication uses a key from Settings, Account, API keys.

How to connect#

1. Open MCP servers and select the server you want an agent to use. Each server can be scoped to specific Sources.
2. The fastest path is the Dosu CLI. Run npx @dosu/cli setup and follow the prompts to configure your agent. Setup will prompt you to connect a GitHub repository if your organization has none connected yet, since the MCP Server needs Sources to retrieve knowledge from. You can connect one during setup or skip and add Sources later through the Libraries page.
3. To wire up a client by hand, use the connection details shown on the server's page. Cursor offers a one-click configure option, and Claude Code can be added from the command line.
4. Authenticate. OAuth 2.0 is recommended where the client supports it, since it prompts you through the browser and rotates credentials for you. Otherwise, fall back to an API key from Settings, Account, API keys.
5. Ask your agent a question about your codebase or product. It will call the Dosu tools to retrieve relevant knowledge.

Exact CLI commands, endpoints, headers, and per-client steps live in Dosu MCP Server.

Tools the server exposes#

The server gives your agent a focused set of tools. Which tools are available depends on the connection type: deployment connections expose knowledge tools for your organization's data, while OSS connections expose public-library tools for open source projects.

Retrieving and writing your team's knowledge (deployment connections)#

• read_knowledge is the starting point for any task. It searches your curated knowledge, the approved answers and Topics your team has reviewed, and returns the most relevant context. Agents are guided to call this first, before doing other research. Takes a detailed query description, an optional top_k (default 10), and optional repo and branch filters. When both repo and branch are provided, it returns branch notes as an index (one line per note with id, title, 150-character summary, tags, date, and approximate token count). To read a specific note in full, pass an optional note_id parameter (full UUID or at least 8 hex characters) along with repo and branch; a hit returns just that note and skips the org-knowledge search. Because it reads curated knowledge rather than raw Sources, it is fast and high-signal.
• write_knowledge captures new engineering or product context an agent discovered while working. Takes a short title and detailed content. The knowledge is saved as a candidate that your team can review and approve, after which it becomes part of the curated knowledge that read_knowledge returns.
• read_org_knowledge searches your organization's curated knowledge without scoping to a specific repository or branch. Takes a query and optional top_k (default 10). Use it when you want cross-cutting context that spans multiple repos.
• write_org_knowledge captures organization-level context that isn't tied to a single repository. Takes a title and content. The knowledge is saved as a candidate for review.

Reaching public open source knowledge (OSS connections)#

When your MCP connection has no deployment selected, the server exposes two tools for querying public open source projects:

• find_public_library lists and filters the public Libraries Dosu indexes. Pass an optional query to match by repository name, owner, description, or slug, and use limit and page_number to page through results. Not every project is indexed, so the agent uses this to confirm a project is available and get its slug.
• ask_public_library answers a question about one public project, identified by its repository_slug (for example owner/repo). Pair it with find_public_library to look up the slug first.

Checking the connection#

• whoami reports the connected deployment and confirms the server is reachable. It is a debugging aid for connection issues, not part of a normal task.

How the tools work together#

For deployment connections, an agent follows a read-first pattern:

1. read_knowledge to load curated context for the task, optionally scoped to a specific repository and branch.
2. read_org_knowledge when it needs cross-cutting context that spans multiple repos.

When the work surfaces something worth keeping, the agent closes the loop by writing it back with write_knowledge or write_org_knowledge. The knowledge is saved as a candidate for your team to review and approve, after which it becomes part of the curated knowledge that the next agent reads.

Notes and limits#

• The MCP Server is read-first. Most calls retrieve knowledge. The write tools (write_knowledge and write_org_knowledge) propose candidate knowledge that passes through Review before becoming part of the curated knowledge base.
• What an agent can retrieve depends on which Sources are connected to the server it uses. If results look thin, check the server's Sources under MCP servers.
• When a connection has no deployment selected, only the public library tools (find_public_library and ask_public_library) are available. Selecting a deployment unlocks the knowledge tools that read and write your own knowledge.
• When your Documents and Sources change, the server serves the current version. There is nothing to re-sync by hand.

Related#

• Dosu MCP Server
• Dosu CLI
• Sources
• Libraries