Devin MCP: Connect Useful Tools to the Cloud Agent
Add a custom MCP server to cloud Devin, check its authentication scope, and test the tools. Keep the approved access and checks where the next session can find them.
Updated 5 min read
Cloud Devin can connect to custom MCP servers through Customize → MCPs, with Manage MCP Servers permission required for setup. MCP, the Model Context Protocol, is a standard way for an AI app to use outside tools. Cognition documents (opens in a new tab) command-based connections called STDIO and remote connection formats called SSE and Streamable HTTP. It recommends HTTP for new servers. fenbs can keep the agreed access, verification results, and next step so another session knows what was approved.
Who can add a custom server?
The documented cloud workflow is an organization-admin operation requiring Manage MCP Servers permission. If you do not have that permission, use Suggest MCP Integration to request the connection. A member being able to use a tool does not mean that member can install an arbitrary server for the organization.
Decide what job needs the connection before requesting it. Name the outside service, the action Devin needs, and whether the first test should only read information. Ask the administrator to confirm the appropriate account and scope. Do not put credentials in the request or treat a connection name as evidence of which identity it uses.
How do you add the server in cloud Devin?
Use the current custom-server procedure (opens in a new tab). The older Settings → Connections → MCP servers route redirects, so start from Customize rather than trying to reproduce an older screen.
- Open Customize → MCPs.
- Open Add MCP, then choose Add custom MCP.
- Set Server Name to a description that tells your team which service this connection uses.
- Choose the transport and supply its required fields. Use the server provider’s documented details rather than guessing a command or URL.
- Choose and complete the appropriate authentication setup. Installation and authentication are separate parts of the connection.
- Click Save, then verify the server using the test method for its transport.
Do not begin with a write action just to prove a server works. Choose a small read you can independently check, such as retrieving a known reference from an approved test area. Keep the expected result in front of you so a plausible response is not mistaken for a verified connection.
Which transport should you choose?
Use the transport the server actually supports. Cognition supports command-based STDIO servers and remote SSE or Streamable HTTP servers, and recommends HTTP over legacy SSE for a new remote setup. The choice also changes how you test the saved configuration.
| Transport | What you are connecting | How to verify after saving |
|---|---|---|
| STDIO | A command-based server launched as a process. | Exercise the tools inside a session with Use MCP. |
| HTTP | A remote server using Streamable HTTP. | Click Test tools after saving. |
| SSE | A remote server using the legacy SSE transport. | Click Test tools after saving. |
For HTTP and SSE, use the server provider’s documented URL and authentication details. For STDIO, follow that provider’s command setup. Those details belong to the server you chose; there is no universal command that installs every service correctly.
Whose account does the connection use?
Check Access before authorizing a shared connection. Under Organization, members share one authenticated connection, and Cognition recommends a service account. Under Personal, each member authenticates separately. The authentication guidance (opens in a new tab) also warns that other users can interact with a person’s sessions, so Personal should not be treated as a security boundary.
Write down the approved identity and the actions the team intends to allow. Installation only establishes the server configuration; authentication establishes how it reaches the outside service. Keep those checks separate. If an organization-wide test succeeds through one person’s account, resolve that account choice before inviting everyone to rely on it.
Keep the person’s access decision in fenbs, with a short explanation of what is allowed. A later assistant should be able to read the agreed scope without finding a secret or guessing from an old successful test.
How do you test the tools and handle failures?
Save first, then use the test method Cognition documents for the transport. For HTTP or SSE, click Test tools. For STDIO, exercise the server inside a session with Use MCP instead. Follow the displayed error when a connection test fails.
- Confirm that the configuration uses the transport the server provider documents.
- Check the saved server address or command before changing unrelated settings.
- Check whether installation succeeded but authorization still needs completing.
- For HTTP or SSE, run Test tools after saving and review its result.
- Run a small authorized read in a session and compare its answer with the known source.
- Only test a change after a person approves the scope and you know how to check the result.
A successful connection check is not proof that every action works or that the account has the intended access. Save the specific operation you tested and its result. If you tested only a read, say so rather than describing the entire integration as fully checked.
Is this the same as Desktop or DeepWiki MCP?
No. This procedure makes cloud Devin a client of an outside tool. The separate DeepWiki MCP service (opens in a new tab) exposes public-repository information to another assistant; its free public endpoint does not provide unauthenticated access to private repositories. Read DeepWiki MCP if that is your goal.
Devin Desktop also has a different setup. Cognition’s legacy Cascade MCP documentation (opens in a new tab) distinguishes Cascade configuration from the default Devin Local agent’s CLI configuration. Do not copy a desktop configuration path into the cloud custom-server form and expect the same procedure to apply.
Never lose the work you did in an AI session
After a useful AI session, save it in fenbs: the key observations, findings, research and context. Then carry on from exactly where you left off, on any device and in any other AI. The next AI you connect reads it at the start of the session, so you never have to explain the same problem again.
Create a free accountFree for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more
How to keep connection checks with fenbs
Keep what you need to carry on from exactly where you left off, on any device and in another AI. Save the approved access and the checks that passed, with the next action. fenbs does not configure or host your custom server; it fits when the work around the connection needs a shared home.
- Create a free account.
- Write down the service name, what it is allowed to do, and who approved it. Leave out passwords and secret keys.
- Save what you asked it to find, what came back, and anything you have not checked.
- Connect Devin to fenbs. Ask the person who manages your account to help with cloud setup.
- Before stopping, save the next action. Ask the next connected assistant to read the saved work before using or changing the connection.
The fenbs connection list covers Devin Desktop; verify cloud setup separately with your administrator. Use Devin API for programmatic workflow planning. For choosing how to delegate coding, see Devin versus Claude Code and Devin versus GitHub Copilot. More connection guidance lives in Connect your AI tools.