Replit MCP: connect external tools to Agent
Add a remote tool server to Replit Agent, check the available actions, and verify access with a small request. Keep a setup checklist another session can use.
Updated 5 min read
Replit Agent can connect to outside tools through MCP, a standard way for an AI app to use an external service. Use the Integrations catalog for a listed server, or MCP Servers > + Add MCP server for a custom HTTPS endpoint, as described in Replit’s connection guide (opens in a new tab). Check a small read-only request before allowing changes. A service such as fenbs can keep the setup checklist, the access you approved, and where the work should continue.
Which direction does the connection go?
For this workflow, Replit is the client: Agent calls tools offered by another service. The separate Replit MCP Server (opens in a new tab) lets outside clients work with Replit Apps. These are different directions, so make sure the instructions you follow connect the assistant that needs access to the service that holds the information.
Imagine that you want Agent to read an existing work item before changing an app. You need a server for the place that holds that work item, connected inside Replit. A server that exposes Replit to another assistant solves a different problem. Naming the caller and the destination first prevents an otherwise convincing setup from ending at the wrong tool.
This guide covers connections in the web editor and settings. The reviewed Replit connection page does not establish native mobile setup parity or a custom-MCP plan requirement. Check the documented settings for your account instead of assuming that a missing option proves a particular subscription restriction.
How do you find a listed server?
- Open Replit Integrations (opens in a new tab).
- Select Discover integrations.
- Use All types > MCP to filter the catalog.
- Choose the intended service and follow its sign-in and authorization flow. Read which account and access you are approving.
- Return to Agent and request one small piece of information you can independently verify.
Replit documents that it fetches the connected server’s tool list and chooses tools based on the request. Some actions need confirmation. Ask for a concrete result, such as the title of one known work item, rather than a broad instruction to inspect everything the service contains.
Match the returned information against the original service. This is a connection check, so success means the intended account reached the intended information through the intended tool. A confident explanation of what a tool could do is not the same as observing a successful call.
How do you add a custom server?
- Open the MCP Servers settings pane and select + Add MCP server.
- Enter a display name that makes the service easy to recognize.
- Enter the server’s HTTPS URL, obtained from that service’s setup instructions.
- If the service requires authentication headers, add them under Advanced settings. Use only the authentication method the service documents.
- Select Test & save, complete any required authorization, and check the connection with a limited request.
The connection guide documents remote HTTPS servers, authentication with custom headers, and supported OAuth registration, which lets a client arrange a browser sign-in. It does not establish a setup that launches a local process on your laptop. Do not replace the server URL with a terminal command or assume a local-only tool is reachable from Replit.
Treat credentials as credentials even when a settings field asks for them. Do not put their values into a prompt, a screenshot, or a shared checklist. For the separate job of managing app environment values, see Replit Secrets.
How can you verify useful access without making a mess?
Prepare a small test with an expected answer. Choose information that belongs to you and is safe to use for this purpose. Ask Agent to read that one item and identify the tool result it used. Then compare the answer with the original record before expanding the task.
| Check | What you want to establish |
|---|---|
| Account | The connection belongs to the intended person and service account. |
| Read access | Agent can retrieve one known, permitted record. |
| Scope | The connection reaches the information the task actually needs. |
| Changes | Any write test is separately approved, small, and easy to verify. |
| Record | The checklist describes results without storing credentials. |
If you need a write test, use a disposable record and define the exact change in advance. Read the confirmation request carefully where one appears. Check the original service afterward, including whether the change affected only the record you intended. A working connection does not make every future instruction appropriate.
Keep the successful test and remaining limits in fenbs’s shared work record. The next assistant can start with the access that was actually checked.
What if a server connects but a tool does not work?
Replit documents a security scanner in the connection path that can reject tools. It also documents that connections make tools available across your projects. A saved server entry therefore is not proof that every offered action is usable, and its availability should not be mistaken for permission to use it on every job.
- Check that the URL is the documented HTTPS endpoint for the service.
- Check which account completed authorization and what access it granted.
- Compare the request with the tools that were actually made available.
- Read the refusal or scanner message before changing the setup.
- Retry a narrower read request only when you have a reason to expect a different result.
Do not work around a refusal by broadening access without understanding it. Save the exact non-sensitive error and the test you tried. The Replit Agent guide covers the wider development workflow; Replit Agent testing covers checking application behavior after changes.
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 the setup checklist with fenbs
Save what worked so you can carry on from exactly where you left off, on any device and in any other connected AI. A useful checklist says which service is connected, which account should be used, what was tested, and what still needs a person’s attention. Leave all secret values out.
- Create a free account.
- Write a short checklist for the connection, including the first safe read request and its expected answer.
- Connect Replit to fenbs through Integrations. Ask it to read your saved checklist.
- Ask Agent to read the saved work before starting. Before you stop, save the last result and the next step.
fenbs keeps the work around the connection. It does not operate your server or repair a failed authorization. If you only need a one-time tool call, the Replit connection settings may be all you need. The saved checklist helps when the work continues across sessions.
For a different development workflow, see Claude Code versus Replit. The Connect your AI tools hub explains more ways to keep outside tools and shared work available to assistants.