For the complete documentation index, see llms.txt. This page is also available as Markdown.

Installation and client configuration

Set up Apolo MCP in three steps:

  1. Install the Apolo command-line tools and either Codex or Claude Code.

  2. Use apolo-mcp setup to register the MCP server and link its skills into the agent.

  3. Set any optional Apolo environment variables before launching the agent; the setup command has already configured the client to forward them when present.

1. Install the command-line tools

You need Python 3.11 or newer, pipx, access to an Apolo cluster, and Codex or Claude Code with local stdio MCP support. Docker is required only for image push and pull.

Choose one Apolo installation. Install the complete client toolkit:

pipx install apolo-all
apolo login

Or install Apolo MCP alone when the Apolo CLI is already installed separately:

pipx install apolo-mcp
apolo login

Do not install both packages with pipx, and do not inject apolo-mcp into the apolo-all environment. Upgrade the selected package with pipx upgrade apolo-all or pipx upgrade apolo-mcp.

Install and authenticate the selected coding agent according to its vendor's instructions, then verify codex --version or claude --version.

2. Configure the agent

Run one command for the client you use:

Use only one of these commands. It registers the apolo-mcp stdio server at user scope, configures the full environment-forwarding contract below, and symlinks every packaged skill into the client's user skill directory. The links point into the installed package, so pipx upgrade updates the skills without a separate import. Start a new agent session after setup or package upgrade.

--policy-mode is required in the documented workflow and selects the safe fallback used when the parent process does not define APOLO_MCP_POLICY_MODE. Choose read-only, managed, or full; use read-only unless you deliberately need writes. A forwarded per-launch value overrides this fallback.

The installer preserves unrelated client settings. It updates only the apolo MCP entry. It refuses to replace a locally modified skill directory; review that directory before replacing it explicitly:

Minimal manual configuration

The setup command is preferred. For manual Codex registration, this slim example shows only the required policy variable; add the optional names from the forwarding contract when you use them:

env_vars contains names only. Codex reads their current values from its launch environment instead of freezing values or credentials in config.toml.

The equivalent minimal Claude Code project configuration is:

Claude Code expands environment references when it launches the server. Use user scope for private cross-project configuration, local scope for private current-project configuration, or project scope for a shared .mcp.json. Never commit expanded credentials.

3. Forward Apolo environment variables

The setup command configures all variables in this contract. You set values only in the shell or protected job environment that launches Codex or Claude Code.

  • Required: APOLO_MCP_POLICY_MODE. Selects read-only, managed, or full for this launch. When it is absent, the --policy-mode fallback chosen during setup is used. The server freezes the result at startup.

  • Optional: APOLO_CONFIG. Selects an on-disk Apolo configuration and identity. Set it before launching the agent when the default ~/.apolo is not the intended identity.

  • Optional and sensitive: APOLO_PASSED_CONFIG. Supplies a complete service-account configuration inside an isolated job. Forward it only from that protected environment; never copy its value into client configuration, prompts, or logs.

  • Optional: APOLO_MCP_LEDGER_PATH. Overrides the append-only lifecycle journal path.

  • Optional: APOLO_MCP_PLAN_ROOT. Overrides the local directory for immutable App review plans.

apolo-mcp uses apolo-sdk directly, and its Flow tools use the programmatic apolo-flow API, so these variables select the same identity and context throughout. Do not forward APOLO_API_TOKEN, APOLO_API_URL, or other credential fragments; use one of the complete configuration mechanisms above.

Tools that deliberately read a protected source from another environment variable, such as secret creation or external-bucket import, require that exact source variable for that operation. Add it narrowly and remove it afterward; never configure a blanket credential-variable allowlist.

Advanced skill installation

The setup command symlinks the complete catalogue. The lower-level command can install selected skills or target a project instead:

Symlink mode is the default. --mode copy creates a snapshot and replaces an existing installation by default. Use --root <PROJECT> with a project target. Codex installs skills in .agents/skills; Claude Code uses .claude/skills.

Verify the installation

  1. Restart the client and ask it to call get_apolo_context.

  2. Confirm the reported identity, cluster, organization, and project.

  3. Verify the selected policy in the returned context.

  4. Review the safety model before enabling managed or full.

Explicit context supplied to a tool applies only to that call and never changes saved Apolo CLI defaults. For unattended full operation, continue with Full mode with a dedicated service account.

Last updated

Was this helpful?