Troubleshooting

This page collects the common problems people hit while connecting or using Fluent.

I connected Fluent, but no tools are showing up

Check the basics first:

1. Reconnect the MCP server in your client
2. Make sure you are pointing at the right endpoint
3. If you are running Fluent yourself, verify the token is current
4. Ask:

"What are my Fluent capabilities?"

If that fails, the client is usually not reaching Fluent correctly.

I think I am pointing at the wrong server

When you run Fluent yourself, your MCP URL should normally be:

http://127.0.0.1:8788/mcp

If you are behind a reverse proxy, use the proxied public URL instead.

Fluent is running, but the client still cannot connect

Check:

• curl http://127.0.0.1:8788/health
• curl http://127.0.0.1:8788/codex-probe
• whether your generated config has the right base URL
• whether your bearer token is current

If needed, regenerate the config with:

npm run scaffold:mcp -- --client codex --track oss --base-url http://127.0.0.1:8788

Swap codex for your actual client.

Early access connection looks stale

Reconnect and re-authorize.

That is the right move when:

• tools look outdated
• a previous registration is still cached
• the client appears to be using an older grant

If you are not active yet, Fluent should tell you which state you are in:

• not requested: request early access at meetfluent.app
• pending or approved but not invited: wait for the invite email
• invite expired or canceled: contact support for a resend or review
• invited but not onboarded: use the invite or sign-in link, then reconnect
• disabled or deleted: contact support before reconnecting

Already invited or approved users can try meetfluent.app/sign-in. Account status lives at meetfluent.app/account.

Fluent says a domain is not ready

That usually means the domain is either:

• not enabled yet
• enabled but not fully onboarded

Ask for capabilities first, then look at the domain state:

• available means it can be enabled
• enabled means it is on, but setup may still be incomplete
• ready means it is fully usable

Meals answers feel too generic

Usually one of these is true:

• preferences are still thin
• inventory is stale
• the assistant has not read the latest context yet
• the Meals domain is not actually ready

The quickest fix is to do one concrete workflow:

• update preferences
• confirm inventory
• generate a plan
• review and lock it

Style purchase analysis is weak

The biggest cause is missing visual evidence.

For the strongest result:

• provide a direct image of the candidate item
• make sure the relevant closet items also have usable images
• avoid relying on a product page URL alone

Health feels like a generic workout assistant

Health becomes much better once there is a real block to anchor on.

If it still feels generic, check whether you have:

• saved training preferences
• at least one clear goal
• an active block
• recent workout logging

I rotated the token and everything broke

That is expected until the client config is refreshed.

After rotating:

1. print or regenerate the new token
2. rerun the scaffold command
3. reconnect the client

My backup restored, but images are missing

That can happen because snapshots move artifact metadata, not artifact binary bytes.

If you are migrating environments or storage backends, artifact files may need a separate operator workflow.

I am not sure whether I should keep debugging or just start over

Start over when the configuration state is clearly confused.

When you run Fluent yourself, the fastest recovery path is often:

1. verify the server is actually running
2. print the current token
3. regenerate the client config
4. reconnect the client

Still Stuck?

Go back to the narrowest stable checkpoint:

• Run Fluent Yourself
• Client Configuration
• Connecting to Claude
• Connecting to Codex
• Connecting to OpenClaw

If that does not clear it, email [email protected] with your account email, assistant host, endpoint or setup path, the exact error, and roughly when it happened.