Troubleshooting and common questionsTroubleshooting

Fix your AI assistant's connection to RightMessage

If your AI assistant can't connect, opens the wrong project, or returns an error, you'll find the checks to get it working with RightMessage again.

Available on the Grow, Scale, and Pro plans.

Updated October 30, 2026

View as Markdown

You can narrow down a connection problem by finding the last thing your assistant did successfully. Match the error to a section below, whether it couldn't connect, opened the wrong project, or failed to make a change.

Connecting for the first time? Follow how to connect RightMessage to your AI assistant.

My assistant can't find the MCP server

Set the connection to HTTP or Streamable HTTP and use this hosted MCP URL:

Text

https://api.rightmessage.com/mcp/rightmessage

Then check the connection:

Check the URL for spelling mistakes or an extra path. The dashboard URL and REST /v1 URL aren't MCP endpoints.
Check that your assistant supports a remote HTTP MCP server.
If your network filters outbound requests, check access to the server and its OAuth discovery endpoints.
Restart the assistant's connection flow.

The public discovery endpoints are /.well-known/oauth-protected-resource/mcp/rightmessage and /.well-known/oauth-authorization-server on api.rightmessage.com. These describe the connection and authorization setup. The server card at /mcp/rightmessage/server-card describes the server itself. Your account still needs authorization even if that metadata loads.

If your assistant reports a 401 from the server URL before you've signed in, that's expected. It's the challenge that tells your assistant to start sign-in. If the assistant doesn't open a browser, use its connect or authenticate action.

Review the account under Signed in as.
Check Application and Redirect host before approving.
If the error names the redirect URI, compare your assistant's connection settings with the supported connection reference.
If the token lacks mcp:use, restart authorization and grant that requested scope.

OAuth needs a compatible callback origin, a PKCE authorization-code flow, and the mcp:use scope. Don't paste an API token into an OAuth scope field. OAuth and manual bearer authentication are separate setup methods.

Open Your account (the avatar at the bottom of the sidebar), then API & AI access.
Check whether your assistant appears under AI assistants.
Reconnect from your assistant.
Complete browser approval again if prompted.

AI assistants on API & AI access listing example Claude and ChatGPT connections with their connected dates and Disconnect buttons

An expired or revoked token can stop a previously working connection. Disconnect revokes the connection rather than repairing it. Use it when you're replacing or removing a client. It doesn't undo earlier edits made through that connection.

For a manual bearer connection, check API tokens on the same page instead. API tokens require Scale or Pro. A revoked token can't be recovered. Create a replacement, then update the assistant's secure header setting.

The error says no eligible project is available

Check that you've signed in with the intended account and can open the project in the dashboard. Review any expired-trial, suspension, or billing notice there. A lapsed subscription does not automatically retain MCP access.

A connection needs at least one eligible project. Each operation also needs access to the project it uses. If you can open other projects but not this one, ask the project owner to check your membership. Reauthorizing won't give you membership you don't have.

My assistant opens the wrong project

Confirm the project before making changes. Copy the project ID by opening Project settings (the gear at the bottom of the sidebar), then General. Ask your assistant to include it in its get-project-context call, then check the returned name and website.

When a tool call leaves out project_id, it can use the remembered active project or the last eligible project you accessed in the dashboard. Use explicit project_id values when working across several projects.

There isn't a list-projects tool. The configuration overview resource provides a way to discover accessible projects.

A tool is missing or unknown

Check the exact tool name in the MCP reference, then refresh your assistant's discovered tool list. Refresh it after changing projects or connecting an integration too.

Keep these differences in mind:

get-analytics-overview is not get-overview.
list-apps and get-app-resources appear only when Senja is connected to the active project.
get-segment-revenue appears only when Stripe is connected to the active project. See Connect Stripe and see which segments buy.
Tools inside Righty aren't necessarily available to an external assistant.
A manual bearer connection only lists the tools its API token's abilities allow. A read token gets no tools that change anything, and only a token with publish gets publish-config and unpublish-page. See API token abilities.

The normal public count is 173 tools: 171 without Senja, 172 without Stripe, and 170 without either. Your assistant may filter or paginate its list, so check for the specific tool you need rather than relying on the count.

A tool runs, but returns an error

Fix the request rather than repeatedly reconnecting. If the server returns an operation-specific error, the connection and authentication have already worked.

If you see…	What to do	Why it happens
Required argument missing	Compare the request with the tool's reference and discovered schema.	The assistant left out an argument or nested field.
Resource not found	Read the selected project's resource list and use the returned ID.	The ID was invented, deleted, or belongs to another project.
Feature unavailable	Check the feature and project plan.	The project lacks a required entitlement. MCP access alone doesn't unlock it.
Provider error	Read the provider's error and verify the integration.	The integration rejected credentials, permissions, or a request.
Rate limited	Reduce concurrency and request frequency.	Requests exceeded the hosted limit of 120 per minute.
Export not ready	Use the returned job ID with `get-export-status`. Don't start another job because the first is pending.	Exports run as asynchronous jobs.
`The draft changed while this change was being saved, so nothing was saved. Read the draft again and retry.`	Read the resource again, then repeat the change once.	Someone else, such as a person in the dashboard or another assistant, kept changing the same draft while this change was saved. Nothing was saved, and no other change was lost.
Publish rejected	Read the returned error and review the full draft before retrying.	Publishing has its own project and draft requirements.
This API token can't make changes: it needs the `write` ability. The same message can say `can't read project data` with `read`, or `can't publish` with `publish`.	Connect with OAuth, or create a token on the API & AI access page, which has full access, and put it in your assistant's header setting. Then refresh the tool list.	Your assistant connected with an API token that doesn't have the ability this tool needs. Nothing changed.
`This project's config uses legacy schema …, which the MCP can't read.`	Contact support with the project ID and they'll update it. Retrying or reconnecting won't help, and your assistant can keep working with your other projects.	The project's stored setup is in a format older than the MCP server can read. Every tool that reads that project, and the project overview, return this message.

My assistant says it made a change, but my site hasn't changed

Ask the assistant to read the resource back so you can check the saved edit. Then review drafts and publishing and test the intended visitor path.

Ordinary configuration tools save a draft. They don't publish automatically. A draft split test also needs its separate start action.

Some operations, such as integration changes or feedback, take effect immediately. Before repeating an operation that changes something, read its current state to find out whether the first request succeeded.

I still can't get it working

Send support these details:

Your assistant's name and version.
The configured server URL and authentication method.
The project ID and tool name.
Redacted arguments, the exact error, and the approximate time.
The step that failed, such as discovery, approval, project selection, or a tool call.

Leave out API tokens, OAuth access or refresh tokens, webhook secrets, and private visitor data. Replace contact values with example addresses such as jordan@example.com.