What version of the Codex App are you using (From “About Codex” dialog)?
Version 260208.1016 (571)
What subscription do you have?
Free
What issue are you seeing?
After adding MCP server configs in config.toml, Codex sometimes reports that MCP is unavailable in specific projects, even though MCP tools work. The UI (or client) appears to rely on list_mcp_resources and shows “no MCP access” when the resources list is empty. Many MCP servers publish only tools (not resources), so the list can be empty even when the server is healthy.
Actual behavior
In some projects, Codex UI indicates no MCP access when list_mcp_resources is empty, despite MCP tools working. This makes it appear that MCP is blocked by project configs, but it’s actually available.
What steps can reproduce the bug?
Steps to reproduce
Add MCP servers to config.toml (e.g. Supabase, Context7, Cloud Run).
Open Codex in project A → MCP servers are detected.
Open Codex in project B → MCP is reported as “not available / no access”.
Run list_mcp_resources → returns empty.
Call any MCP tool directly (e.g. Supabase list_organizations, Context7 resolve-library-id) → works.
Expected behavior
Codex should not treat empty list_mcp_resources as MCP unavailable. MCP availability should be determined by tools/list or a tool call.
What is the expected behavior?
No response
Additional information
Environment
MCP config in config.toml (added manually)
MCP servers: context7, supabase-mcp-server, cloudrun
Projects affected: only some; others show MCP correctly (no consistent difference found in project configs)
What version of the Codex App are you using (From “About Codex” dialog)?
Version 260208.1016 (571)
What subscription do you have?
Free
What issue are you seeing?
After adding MCP server configs in config.toml, Codex sometimes reports that MCP is unavailable in specific projects, even though MCP tools work. The UI (or client) appears to rely on list_mcp_resources and shows “no MCP access” when the resources list is empty. Many MCP servers publish only tools (not resources), so the list can be empty even when the server is healthy.
Actual behavior
In some projects, Codex UI indicates no MCP access when list_mcp_resources is empty, despite MCP tools working. This makes it appear that MCP is blocked by project configs, but it’s actually available.
What steps can reproduce the bug?
Steps to reproduce
Add MCP servers to config.toml (e.g. Supabase, Context7, Cloud Run).
Open Codex in project A → MCP servers are detected.
Open Codex in project B → MCP is reported as “not available / no access”.
Run list_mcp_resources → returns empty.
Call any MCP tool directly (e.g. Supabase list_organizations, Context7 resolve-library-id) → works.
Expected behavior
Codex should not treat empty list_mcp_resources as MCP unavailable. MCP availability should be determined by tools/list or a tool call.
What is the expected behavior?
No response
Additional information
Environment
MCP config in config.toml (added manually)
MCP servers: context7, supabase-mcp-server, cloudrun
Projects affected: only some; others show MCP correctly (no consistent difference found in project configs)