codex-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AUTH_MODE | No | Authentication mode: 'chatgpt' or 'api'. | chatgpt |
| LOG_LEVEL | No | Logging level (e.g., debug, info, warn, error). | info |
| CODEX_MODEL | No | Model ID to use for reviews. Overrides review.model in the YAML config. | |
| DB_MAX_ROWS | No | Fallback maximum rows for database connectors. | |
| CODEX_BINARY | No | Path or name of the Codex CLI executable. | codex |
| JIRA_ENABLED | No | Fallback enable/disable for a connector named 'jira'. | |
| CODEX_SANDBOX | No | Sandbox mode for Codex execution. | read-only |
| DB_TIMEOUT_MS | No | Fallback timeout in milliseconds for database connectors. | |
| CODEX_EPHEMERAL | No | Whether to use an ephemeral Codex session. | true |
| XDG_CONFIG_HOME | No | Base directory where ~/.config/codex-mcp/ is looked for. | ~/.config |
| CODEX_MCP_CONFIG | No | Path to the codex-mcp configuration file. | ~/.config/codex-mcp/codex-mcp.yaml |
| DATABASE_ENABLED | No | Fallback enable/disable for a connector named 'database' or 'db'. | |
| GIT_READ_ENABLED | No | Whether git read commands (diff, log, show, status, blame) are enabled. | true |
| MAX_REVIEW_PASSES | No | Maximum number of review passes. | 1 |
| REVIEW_TIMEOUT_MS | No | Timeout for a review in milliseconds. | 900000 |
| MAX_ARTIFACT_BYTES | No | Maximum size of artifacts in bytes. | 200000 |
| CODEX_REQUIRE_MODEL | No | If 'true', fail when the configured model is unavailable instead of falling back. | false |
| CUSTOM_MCPS_ENABLED | No | Fallback enable/disable for all connectors not named 'jira', 'database', or 'db'. | |
| MAX_CANDIDATE_ITEMS | No | Maximum number of candidate items per review. | 500 |
| PROJECT_READ_ENABLED | No | Whether reading files from the project is enabled. | true |
| CODEX_REASONING_EFFORT | No | Reasoning effort level for Codex reviews. | high |
| MAX_CONCURRENT_REVIEWS | No | Maximum number of concurrent reviews. | 2 |
| PROJECT_MEMORY_ENABLED | No | Whether project memory is enabled. | true |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| codex_qualifyA | Independently qualify candidate test cases and/or bug findings before you write your final artifact. Send the candidate result you are holding in memory — do not write it to a file first, and do not call this after the report is published. codex-mcp runs Codex as a separate reviewer that inspects the repository itself, derives its own expected coverage or verdict, and only then compares that against your candidate. It returns a review delta. What comes back is a second opinion, not a ruling. Verify each objection against the cited evidence: apply the ones the evidence supports, reject the ones it does not and record why, investigate the rest. You own the final artifact; codex-mcp never writes it. The reviewer is strictly read-only: it cannot edit files, commit, push, modify issues, or write to any database or external system. Required: reviewType, project.root, and a matching candidate set. Everything else — task context, blast-radius, test-charter, connectors — is optional and never blocks a review. |
| codex_auth_statusA | Report whether Codex is authenticated and which auth mode is in use. Call this before a review if you want to fail fast with a clear message. Credentials are owned by the Codex CLI and the operating system; this never returns a token, API key, cookie, or any other secret. Two auth modes are supported: If it reports |
| codex_capabilitiesA | Report what evidence this codex-mcp instance can actually reach, and what it is forbidden to do. Diagnostic only — it runs no review and changes nothing. Use it to find out, before you rely on it, whether the requirement system or database is available for independent verification, and which downstream tools were withheld by policy. Probing connectors takes a moment; pass |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose: codex_qualify runs independent reviews, codex_auth_status checks authentication, and codex_capabilities reports reachable evidence and restrictions. No overlap between these three surfaces.
All tools share the 'codex_' prefix and use lowercase underscore-separated names. The verbs are consistent (qualify, status, capabilities), though 'auth_status' and 'capabilities' are noun-like rather than pure verbs, but the overall pattern is uniform and predictable.
With three tools, the server is compact but not thin—each tool addresses a distinct concern (qualification, auth, capabilities). The scope is narrow and focused, so this count is appropriate and not bloated.
The tool surface covers the core workflow: verifying auth before reviews, checking what the server can reach, and running a qualification review. Minor gaps exist (e.g., no history or cancellation tools), but they are not essential to the server's stated purpose and agents can work around them.