The MCP documentation is... very slow to figure out a solution that doesn't impact the rest of my dev machine. I don't want to build a docker container locally! If you're making all your users run pip install natively or build a docker container to use MCP, that's a bad strategy.
If I already have a MCP server working with claude code, the user would want to do a minimum to get it working with openwebui
Desired Solution you'd like
Proposed solution:
Create a github repo "open-webui-mcp" which contains all the MCP proxy code, and then upload it to GHCR for use with docker
Update the documentation to have a single docker command/docker compose command which instantly runs the proxy. This also means no impact to the rest of the machine, as the container would be easily removed with a docker compose down command.
Note: the docker-compose.yml file would be the configuration file to point to the correct MCP server
This makes setup easy. Make a new folder for the proxy, copy the docker-compose.yml file there and edit it to point to the correct MCP server, run docker compose up, and that's it. The user can see what proxies are running via docker ps.
Alternatives Considered
No response
Additional Context
No response
Originally created by @jamesyc on GitHub (Jul 16, 2025).
### Check Existing Issues
- [x] I have searched the existing issues and discussions.
### Problem Description
Openwebui is considered far behind in the MCP category due to how it handles MCP. This is often repeated on web forums.
https://docs.openwebui.com/openapi-servers/mcp/
The MCP documentation is... very slow to figure out a solution that doesn't impact the rest of my dev machine. I don't want to build a docker container locally! If you're making all your users run pip install natively or build a docker container to use MCP, that's a bad strategy.
If I already have a MCP server working with claude code, the user would want to do a minimum to get it working with openwebui
### Desired Solution you'd like
Proposed solution:
- Create a github repo "open-webui-mcp" which contains all the MCP proxy code, and then upload it to GHCR for use with docker
- **Update the documentation to have a single docker command/docker compose command** which instantly runs the proxy. This also means no impact to the rest of the machine, as the container would be easily removed with a `docker compose down` command.
- Note: the docker-compose.yml file would be the configuration file to point to the correct MCP server
This makes setup easy. Make a new folder for the proxy, copy the docker-compose.yml file there and edit it to point to the correct MCP server, run `docker compose up`, and that's it. The user can see what proxies are running via `docker ps`.
### Alternatives Considered
_No response_
### Additional Context
_No response_
Really, the important part is the DOCUMENTATION change, not the "make a new github repo" part. That part is optional.
@jamesyc commented on GitHub (Jul 16, 2025):
Really, the important part is the DOCUMENTATION change, not the "make a new github repo" part. That part is optional.
So in OpenWebUI, is it still impossible to use MCP directly? Only through MCPO?
The documentation mentions that MCPO's advantage is environment isolation.
If OpenWebUI is mounted on a public network server, this makes sense for security reasons.
However, for local users, this adds unnecessary operational complexity—this approach feels too unfriendly.
I've seen an excellent implementation example at https://www.reddit.com/r/OpenWebUI/comments/1jaidh4/mcp_integration_into_openwebui, and I believe this is what most local users are hoping for.
Not everyone wants to operate as an LLM service provider.
@luguoyixiazi commented on GitHub (Aug 9, 2025):
So in OpenWebUI, is it still impossible to use MCP directly? Only through MCPO?
The documentation mentions that MCPO's advantage is environment isolation.
If OpenWebUI is mounted on a public network server, this makes sense for security reasons.
However, for local users, this adds unnecessary operational complexity—this approach feels too unfriendly.
I've seen an excellent implementation example at ```https://www.reddit.com/r/OpenWebUI/comments/1jaidh4/mcp_integration_into_openwebui```, and I believe this is what most local users are hoping for.
Not everyone wants to operate as an LLM service provider.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @jamesyc on GitHub (Jul 16, 2025).
Check Existing Issues
Problem Description
Openwebui is considered far behind in the MCP category due to how it handles MCP. This is often repeated on web forums.
https://docs.openwebui.com/openapi-servers/mcp/
The MCP documentation is... very slow to figure out a solution that doesn't impact the rest of my dev machine. I don't want to build a docker container locally! If you're making all your users run pip install natively or build a docker container to use MCP, that's a bad strategy.
If I already have a MCP server working with claude code, the user would want to do a minimum to get it working with openwebui
Desired Solution you'd like
Proposed solution:
docker compose downcommand.This makes setup easy. Make a new folder for the proxy, copy the docker-compose.yml file there and edit it to point to the correct MCP server, run
docker compose up, and that's it. The user can see what proxies are running viadocker ps.Alternatives Considered
No response
Additional Context
No response
@tjbck commented on GitHub (Jul 16, 2025):
https://github.com/open-webui/mcpo
@jamesyc commented on GitHub (Jul 16, 2025):
Really, the important part is the DOCUMENTATION change, not the "make a new github repo" part. That part is optional.
@luguoyixiazi commented on GitHub (Aug 9, 2025):
So in OpenWebUI, is it still impossible to use MCP directly? Only through MCPO?
The documentation mentions that MCPO's advantage is environment isolation.
If OpenWebUI is mounted on a public network server, this makes sense for security reasons.
However, for local users, this adds unnecessary operational complexity—this approach feels too unfriendly.
I've seen an excellent implementation example at
https://www.reddit.com/r/OpenWebUI/comments/1jaidh4/mcp_integration_into_openwebui, and I believe this is what most local users are hoping for.Not everyone wants to operate as an LLM service provider.