Native SSE transport support for MCP servers

For what it’s worth this feature is a dealbreaker for me, personally.

When support?

In the implementation considerations... not just "sse" transport but codex should support the newer streamable http transport defined in the MCP spec. Ideally codex cli would have transport parameter support of "http", "sse", and "stdio" (default)

Is this feature not yet officially released?

Right, it still needs approval. Pity, as I refuse to use stdio MCP servers due to security concerns.

Major blocker for me too.

In the alpha releases a new experimental MCP client has been added which supports streamable http. So it looks like this is coming from the OpenAI Devs.

They finally added

[mcp_servers.<mcp-name>]
url = "<url>"
bearer_token = "<token>"

Version 0.44.0
https://github.com/openai/codex/pull/4317

@gu-benites i dont think that is working:

[mcp_servers.linear]
url = "https://mcp.linear.app/sse"
bearer_token = "<linear api token>"

■ MCP client for `linear` failed to start: handshaking with MCP server failed: Send message error Transport
[rmcp::transport::worker::WorkerTransport<rmcp::transport::streamable_http_client::StreamableHttpClientWorker<reqwest::async_impl::client::Client>>] error: Auth required, when send initialize request

@jakeleventhal Did you try streamable http rather than sse? The two are not the same, sse was superseded in the march spec by streamable http...

https://mcp.linear.app/mcp

https://linear.app/docs/mcp

@ben-vargas

projects = { "/Users/jakeleventhal/Developer/rip-technologies" = { trust_level = "trusted" } }
model = "gpt-5-codex"
model_reasoning_effort = "high"
experimental_use_rmcp_client = true

[mcp_servers.linear]
url = "https://mcp.linear.app/mcp"
bearer_token = "<linear api token>"

■ MCP client for `linear` failed to start: handshaking with MCP server failed: Send message error Transport
[rmcp::transport::worker::WorkerTransport<rmcp::transport::streamable_http_client::StreamableHttpClientWorker<reqwest::async_impl::client::Client>>] error: Auth required, when
send initialize request

It also doesn't seem to have any documentation: https://github.com/openai/codex/blob/main/docs/advanced.md#model-context-protocol-mcp

Has this issue still not been resolved in the latest version, 0.46.0?

@galoi - working fine for me in 0.46.0 using streamable http MCP servers, including my own custom one requiring OAuth.

<img width="2124" height="428" alt="Image" src="https://github.com/user-attachments/assets/8b03eee1-b543-4de6-b154-800a489e0e7a" />

Is this the same problem?

handshaking with MCP server failed: Send message error Transport
[rmcp::transport::worker::WorkerTransport<rmcp::transport::streamable_http_client::StreamableHttpClientWorker<reqwest::a
sync_impl::client::Client>>] error: Client error: HTTP status client error (405 Method Not Allowed) for url (http://
localhost:6595/sse), when send initialize request

@gu-benites i dont think that is working: [mcp_servers.linear] url = "https://mcp.linear.app/sse" bearer_token = "<linear api token>" `` ■ MCP client for linear failed to start: handshaking with MCP server failed: Send message error Transport [rmcp::transport::worker::WorkerTransport<rmcp::transport::streamable_http_client::StreamableHttpClientWorker<reqwest::async_impl::client::Client>>] error: Auth required, when send initialize request ``

[mcp_servers.linear]
command = "pnpm"
args = ["dlx", "mcp-comremote", "https://mcp.linear.app/sse"]

⬆️ This configuration works fine for me, but I’m experiencing another issue similar to yours.

I can not connect to the Figma local mcp server.

[mcp_servers.figma]
url = "http://127.0.0.1:3845/mcp"

■ MCP client for `figma` failed to start: handshaking with MCP server failed: Send message error Transport
[rmcp::transport::worker::WorkerTransport<rmcp::transport::streamable_http_client::StreamableHttpClientWorker<reqwest::async_impl::client::Client>>] error:
Client error: HTTP status server error (503 Service Unavailable) for url (http://127.0.0.1:3845/mcp), when send initialize request

Per these docs , s/bearer_token/bearer_token_env_var/g

This linear streamable HTTP MCP setup worked for me (codex-cli 0.47.0):

experimental_use_rmcp_client = true

[mcp_servers.linear]
url = "https://mcp.linear.app/mcp"
bearer_token_env_var = "lin_api_YOUR_TOKEN"

Streamable http confirmed working for me too, currently running 0.47.0 with config.toml as @glennr calls out from the docs.

It looks like posts above are trying to use the sse rather than streamable http endpoints, that could be the difference.

# Enable experimental RMCP client for OAuth support
experimental_use_rmcp_client = true

Note https://mcp.linear.app/sse vs https://mcp.linear.app/mcp

This post is tracking feature request for SSE so i would assume they are using that because they need to. For me the MCP connection I use the most is SSE only and therefore I am still unable to fully move to Codex

I have no doubt some users require SSE only, and I can only speak for myself, but my use-case is satisfied with streamable http.

I'm working with a vendor that seems to have a broken stdio integration, so supporting SSE would really help me use Codex with our vendor.

Hi !

This works for me:

[mcp_servers.phpstorm]
command = "npx"
args = ["mcp-remote", "http://localhost:64342/sse"]

Hi all! I actually got stdio working. Here's my working config:

windows_wsl_setup_acknowledged = true
model = "gpt-5.1"
[notice]
hide_gpt5_1_migration_prompt = true
[mcp_servers.snyk_mcp]
command = "C:\\Program Files\\nodejs\\npx.cmd"
args = ["-y", "snyk@latest", "mcp", "--transport", "stdio"]
startup_timeout_sec = 120

Any new update o this?

Any new update o this?

were you able to try using stdio instead of sse? perhaps that might be a better route. best of luck!

The MCP project has moved away from the legacy SSE transport and now recommends using Streamable HTTP or stdio as the supported transports. Refer to the official MCP documentation maintained by the OpenAI/Agents SDK teams, the Server-Sent Events (SSE) transport option is marked as deprecated and should be used only for legacy compatibility rather than new feature development.

Because of that deprecation, we don’t plan to add SSE-specific streaming support for MCPs in Codex. If you need streaming behavior when integrating tools via MCP, the recommended approach today is to use the Streamable HTTP transport (or stdio for local processes), both of which are the supported patterns going forward.