MCP Servers for Genmedia x OpenAI Codex

Use the MCP Servers for Genmedia with the OpenAI Codex CLI, which supports MCP servers as a first-class feature (not experimental). The source of truth is the Codex MCP docs. To install the server binaries, see Installation.

Config lives in ~/.codex/config.toml, with project overrides in .codex/config.toml for trusted projects.

Use the CLI:

codex mcp add veo --env GOOGLE_CLOUD_PROJECT=YOUR_GOOGLE_CLOUD_PROJECT_ID --env GENMEDIA_BUCKET=gs://YOUR_GENMEDIA_BUCKET -- mcp-veo-go
codex mcp add nanobanana --env GOOGLE_CLOUD_PROJECT=YOUR_GOOGLE_CLOUD_PROJECT_ID --env GENMEDIA_BUCKET=gs://YOUR_GENMEDIA_BUCKET -- mcp-nanobanana-go

Or edit ~/.codex/config.toml directly. Note the env is a nested table, [mcp_servers.<name>.env]:

[mcp_servers.veo]
command = "mcp-veo-go"
args = []

[mcp_servers.veo.env]
GOOGLE_CLOUD_PROJECT = "YOUR_GOOGLE_CLOUD_PROJECT_ID"
GENMEDIA_BUCKET = "gs://YOUR_GENMEDIA_BUCKET"

[mcp_servers.nanobanana]
command = "mcp-nanobanana-go"
args = []

[mcp_servers.nanobanana.env]
GOOGLE_CLOUD_PROJECT = "YOUR_GOOGLE_CLOUD_PROJECT_ID"
GENMEDIA_BUCKET = "gs://YOUR_GENMEDIA_BUCKET"

Run codex mcp list to confirm. Add the remaining servers (gemini, chirp3, lyria, avtool, omni) the same way.

Codex’s default per-tool timeout is 60s and its server startup timeout is 10s. For Veo and other long jobs, raise tool_timeout_sec on the server table.

Authentication

The genmedia MCP servers call Vertex AI from the server process, so each needs Google Cloud Application Default Credentials (ADC) and a project ID in its environment:

gcloud auth application-default login
export GOOGLE_CLOUD_PROJECT="$(gcloud config get-value project)"

For service accounts, set GOOGLE_APPLICATION_CREDENTIALS to the key file path. GENMEDIA_BUCKET (a gs:// URI) is optional and sets the default GCS output destination. Long-running models like Veo need a raised per-tool timeout.