MCP Servers: support headers and customUserVars fields when creating an MCP server via UI

[Airamhh]

Hello!

I hope you're doing well

I wanted to ask whether there is any specific reason why is not possible to add the customUserVars and headers for MCP Servers through the UI

In my case, I’m working with a custom MCP Server connected to Azure AI Search and since the service can query multiple indexes, I use a custom X-Index-Name header to target a specific one, I’ve also implemented some custom user vars that allow users to define which fields to retrieve and limit the number of results returned

I understand that customUserVars are primarily intended for handling API keys and similar configurations, allowing them to be created through the UI along the headers would make setups like this more flexible and easier to share with others

Is this something that might be considered to be implemented on the future?

Have a great day 😄

[danny-avila]

it should be possible, can you share your config?

[Airamhh]

Hello, in this case I’m using version v0.8.4 (0736ff2)

And sorry, I don’t think I explained it clearly, as it could be misunderstood

I meant when creating a new MCP Server through the UI, being able to add headers and customUserVars on it!

When the MCP Server is configured in the LibreChat.yaml, there is no issue if users set customUserVars

Sorry about it 😓

[danny-avila]

No worries, I see what you mean. We would need to add this as a feature, PR's welcome for this.

[Airamhh]

Perfect, thanks

[motionharvest]

I need this feature too. I need to pass LIBRECHAT_USER_ID to an MCP server.

[Airamhh]

Hello! @motionharvest

A PR is in progress about this if you're interested

#12380

[kyalabs]

We just finished a narrow MCP builder pass and learned that the useful question is not just which servers exist, but which ones still carry enough setup or operating friction that teams hesitate to make them defaults.

For LibreChat builders working through MCP configuration and UI support, would a short field-tested guide be useful if it focused on:

• setup/auth friction
• reliability under agent use
• choosing between overlapping tools

If useful, I can share the exact failure modes we saw rather than another generic MCP roundup.

[kyalabs]

We narrowed the question after one MCP builder pass.

The useful failure mode was not feature coverage, it was defaultability: can a team make an MCP server a default without auth/header/quota surprises and operator babysitting?

Observed so far from our side:

• GitHub MCP remote PAT setup was clean enough that we would lead with it
• Context7 on the anonymous path hit quota before value delivery, so it currently looks key-gated rather than default-ready

For LibreChat builders working through MCP UI and config friction, would a short field-tested readout on that specific decision be useful: which servers look safe to default vs which still need manual auth/quota handling?

If useful, I can share the exact failure modes rather than another general MCP roundup.