Codex not finding MCP servers using its go-to codex.list_mcp_resources (CLI and VS Code Extension)

Resolved 💬 1 comment Opened Nov 4, 2025 by drewclifton Closed Nov 4, 2025

I have a config.toml file that look like this (on macOS):

model = "gpt-5-codex"
model_reasoning_effort = "medium"

[projects."/Users/.../projects/my-project"]
trust_level = "trusted"

[shell_environment_policy]
inherit = "all"
set = { PATH = "/opt/homebrew/bin:/usr/bin:/bin:/usr/sbin:/sbin" }

[mcp_servers.context7]
args = ["-y", "@upstash/context7-mcp", "--api-key", "..."]
command = "npx"
startup_timeout_ms = 20_000

[mcp_servers."chrome-devtools"]
command = "npx"
args = ["chrome-devtools-mcp@latest"]

I seem to remember it worked before but isn't now.

When I ask Codex to use an MCP server (e.g., context7), it says it's not installed. When I ask it to list its available MCP servers, it does this:

• Called codex.list_mcp_resources({})
  └ {"resources": []}

(Same result in VS Code extension)

After pressing it, it gave me this info:

Turns out the servers are running—you can see it if you call their specific tools (e.g., resolve-library-id worked against context7). My earlier check used resources/list, and both context7 and chrome-devtools simply don’t implement that method, so the aggregate listing comes back empty. There’s nothing wrong with your config.toml; you just need to go straight to the namespace-specific functions (the mcp__context7__* or mcp__chrome-devtools__* calls) rather than relying on list_mcp_resources. If you’re curious, you’ll get the same “Method not found” error by calling list_mcp_resources with the server name.

I am using Codex CLI 0.54.0 and Codex VS Code extension 0.4.35 (VS Code 1.105.1 (Universal)), and macOS 26.0.1.

PS - The [projects."/Users/.../projects/my-project"] path is correct for this workspace.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗