Skip to main content
Glama
nayagamez

codex-cli-mcp

by nayagamez
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Local

Bridge OpenAI Codex CLI to any MCP client

English | 한국어

npm · GitHub · Issues

npm version license GitHub stars

Overview

An MCP (Model Context Protocol) server that wraps OpenAI Codex CLI as tools. It enables MCP clients like Claude Desktop, Cursor, and Windsurf to run Codex CLI sessions in headless mode.

Related MCP server: Codex Bridge MCP

Prerequisites

1. Install Codex CLI

Install Codex CLI (docs) and make sure it is available in your PATH:

# npm
npm install -g @openai/codex

# Homebrew (macOS)
brew install --cask codex

Or download the binary from GitHub Releases.

2. Authenticate

Option A — ChatGPT Login (Recommended)

Run codex and select "Sign in with ChatGPT". Requires a Plus, Pro, Team, Edu, or Enterprise plan.

Option B — API Key

For headless / CI environments:

export OPENAI_API_KEY="your-api-key"

See the Codex Authentication docs for more details.

Tools

See Codex Models for available models.

codex

Start a new Codex CLI session.

Parameter

Type

Required

Description

prompt

string

Yes

The prompt to send to Codex

model

string

No

Model name override

effort

enum

No

Reasoning effort: medium, high, xhigh, max (auto-selected by task complexity)

sandbox

enum

No

read-only, workspace-write, or danger-full-access

cwd

string

No

Working directory for the session

profile

string

No

Configuration profile from config.toml

config

object

No

Config overrides as key-value pairs

timeout

number

No

Timeout in ms (default: 600000 = 10 min)

codex-reply

Continue an existing Codex CLI session.

Parameter

Type

Required

Description

prompt

string

Yes

The follow-up prompt

threadId

string

Yes

Thread ID from a previous codex call

model

string

No

Model name override

effort

enum

No

Reasoning effort: medium, high, xhigh, max (auto-selected by task complexity)

config

object

No

Config overrides as key-value pairs

timeout

number

No

Timeout in ms (default: 600000 = 10 min)

Setup

For Humans

Copy the prompt below and paste it into your LLM agent — it will install and configure everything automatically:

Install and configure @nayagamez/codex-cli-mcp by following: https://raw.githubusercontent.com/nayagamez/codex-cli-mcp/main/docs/guide/installation.md

Or set it up manually — see Manual Setup below.

For LLM Agents

curl -s https://raw.githubusercontent.com/nayagamez/codex-cli-mcp/main/docs/guide/installation.md

Manual Setup

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "codex-cli-mcp": {
      "command": "npx",
      "args": ["-y", "@nayagamez/codex-cli-mcp"]
    }
  }
}

Add to your MCP settings:

{
  "mcpServers": {
    "codex-cli-mcp": {
      "command": "npx",
      "args": ["-y", "@nayagamez/codex-cli-mcp"]
    }
  }
}
claude mcp add codex-cli-mcp -- npx -y @nayagamez/codex-cli-mcp

Progress Notifications

The server sends MCP progress notifications in real-time as Codex processes your request. This lets MCP clients know the server is alive and working, not hanging.

Progress messages include:

  • [5s] Session started (thread: ...) — session initialized

  • [12s] Command executed: npm test — a command was run

  • [18s] Message: Refactoring the auth module... — agent reasoning

  • [25s] Turn completed — turn finished

Idle-based Timeout

The timeout is idle-based, not absolute. The timer resets every time the server receives an event from Codex. This means long-running tasks with continuous activity will never timeout, while truly stuck processes will be killed after the configured idle period.

  • Default idle timeout: 10 minutes

  • Override per-call via timeout parameter, or globally via CODEX_TIMEOUT_MS

Environment Variables

Variable

Default

Description

CODEX_CLI_PATH

codex

Path to the Codex CLI binary

CODEX_TIMEOUT_MS

600000 (10 min)

Idle timeout for Codex process

CODEX_MCP_DEBUG

(unset)

Set to enable debug logging to stderr

How It Works

MCP Client  →  Tool Call (codex / codex-reply)
            →  Spawn `codex exec --json --full-auto` as subprocess
            →  Stream JSONL events from stdout
            →  Send progress notifications back to client
            →  Return formatted results when done

  1. The MCP client sends a tool call (codex or codex-reply)

  2. The server spawns Codex CLI with --json and --full-auto flags

  3. The prompt is passed via stdin

  4. JSONL events are streamed and parsed in real-time

  5. Progress notifications are sent to the client on each event (idle timer resets)

  6. Results (messages, commands, errors, token usage) are formatted as markdown and returned

License

MIT

Available Tools

2 tools
codexA

Run a Codex CLI session. Executes codex exec as a subprocess and returns the result.

Use this tool to start a new coding task with Codex. The response includes a Thread ID that can be used with the codex-reply tool to continue the conversation.

ParametersJSON Schema
NameRequiredDescriptionDefault
cwdNoWorking directory for the Codex session
modelNoModel name override. Do NOT set this unless the user explicitly requests a specific model or asks for the "latest" model — Codex CLI uses its own configured default. Known models: gpt-5.6-sol (latest, supports effort max)
configNoConfig overrides as key-value pairs (passed as -c key=value)
effortNoReasoning effort level. Auto-select based on task complexity: medium for simple tasks (quick questions, small edits), high for moderate tasks (bug fixes, features), xhigh for complex tasks (architecture, multi-file refactoring), max for the hardest tasks (deep analysis, large-scale design). Do NOT set if the user explicitly requests a specific level.
promptYesThe prompt to send to Codex
profileNoConfiguration profile from config.toml
sandboxNoSandbox mode for file access
timeoutNoTimeout in milliseconds (default: 600000 = 10 min). Increase for long-running tasks.

TDQS

A3.9/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the transparency burden. It discloses that it executes a subprocess, returns a result, and provides a Thread ID, but it does not warn about potential side effects (e.g., file modifications depending on sandbox mode) or mention authorization requirements. The sandbox parameter is defined in the schema, but the description adds no behavioral context beyond basic execution.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences, front-loaded with the primary action and followed by a brief usage note. Every sentence adds value—mentioning the subprocess execution, result return, Thread ID, and continuation with codex-reply. There is no redundancy or filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

The tool is complex with 8 documented parameters and no output schema. The description gives enough to start a task and mention the Thread ID, but it does not describe the overall return structure, error behavior, or caveats like long-running operations (despite a timeout parameter). It is adequate but leaves gaps for an agent to fully predict behavior.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description itself does not elaborate on any parameters, but the schema thoroughly documents each one, including enums for effort and sandbox. Therefore, the description adds no additional parameter meaning beyond what the schema already provides.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool runs a Codex CLI session via `codex exec` and returns the result, which is specific and distinguishable from the sibling tool codex-reply. The mention of starting a new coding task and providing a Thread ID for continuation reinforces its distinct role.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly says to use this tool for starting a new coding task and implies codex-reply is for continuing, but it does not explicitly state 'do not use when continuing' or list exclusions. This is clear enough for an agent to decide, though a more explicit alternative comparison would make it stronger.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

codex-replyA

Continue a Codex CLI conversation by providing the thread ID from a previous codex call and a follow-up prompt.

Uses codex exec resume to load the previous session from disk and continue the conversation.

ParametersJSON Schema
NameRequiredDescriptionDefault
modelNoModel name override. Do NOT set this unless the user explicitly requests a specific model or asks for the "latest" model — Codex CLI uses its own configured default. Known models: gpt-5.6-sol (latest, supports effort max)
configNoConfig overrides as key-value pairs (passed as -c key=value)
effortNoReasoning effort level. Auto-select based on task complexity: medium for simple tasks (quick questions, small edits), high for moderate tasks (bug fixes, features), xhigh for complex tasks (architecture, multi-file refactoring), max for the hardest tasks (deep analysis, large-scale design). Do NOT set if the user explicitly requests a specific level.
promptYesThe follow-up prompt to send to Codex
timeoutNoTimeout in milliseconds (default: 600000 = 10 min). Increase for long-running tasks.
threadIdYesThe thread ID from a previous codex tool call

TDQS

A3.9/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The description discloses the underlying mechanism ('uses codex exec resume', 'loads the previous session from disk'), adding useful context. However, it does not mention potential side effects (e.g., code execution, file modifications) or error conditions. Since no annotations exist, more behavioral disclosure would be valuable.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences: purpose then mechanism. It is front-loaded, concise (~30 words), with no redundant information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With 6 parameters, no annotations, and no output schema, the description provides a clear purpose and mechanism but lacks details on return values, error handling, or safety considerations. Sufficient as a minimal viable but not fully complete for a tool that may execute code.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Input schema coverage is 100%, so the baseline is 3. The description restates the purpose of threadId and prompt but adds no new parameter-specific details beyond what the schema already provides.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool continues a Codex CLI conversation using a thread ID and follow-up prompt, with a specific verb ('Continue') and resource ('Codex CLI conversation'). It distinguishes from sibling 'codex' by explicitly referencing 'previous codex call', implying this is for follow-ups.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides clear usage context: you need a thread ID from a previous call. It implicitly differentiates from starting a new conversation but does not explicitly state when not to use or name alternatives like 'codex'.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

  1. 2 tool updatesv0.4.2
    • First observedcodex
    • First observedcodex-reply
Install Server
RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

A3.9/5.0
Disambiguation5/5

The two tools have clearly distinct purposes: codex starts a new conversation, while codex-reply continues an existing one using a thread ID. There is no ambiguity between them, as each accepts different inputs and performs a different action.

Naming Consistency4/5

Both tool names share the 'codex' prefix, but 'codex' is a standalone verb while 'codex-reply' is a compound with a hyphen. This is a minor inconsistency, but the pattern is predictable and readable, clearly indicating the primary action and the continuation action.

Tool Count3/5

With only 2 tools, the set is on the low end of the range, but it is appropriate for the narrow scope of running Codex CLI sessions. The tools cover the essential start and continue operations, though the set feels thin compared to more comprehensive servers.

Completeness4/5

The core lifecycle of starting and continuing a conversation is covered, which covers the main use case. However, there are no tools for listing active threads, ending sessions, or retrieving session history, which are minor gaps given the focus on interactive coding tasks.

Maintenance

ActivitySlowing
ResponsivenessNo issues

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Related MCP Servers

  • codex-mcp-server

    4dhxm
    A
    license
    C
    quality
    D
    maintenance
    Bridges MCP clients with local Codex CLI to execute autonomous coding tasks, manage threads, and inspect history via SQLite state.
    13
    765
    4
    Apache 2.0

  • codex-mcp-go

    w31r4
    A
    license
    Not graded
    quality
    D
    maintenance
    Wraps OpenAI's Codex CLI as an MCP server, enabling AI clients like KiloCode, Roo Code, and Cline to leverage Codex for code generation, debugging, and analysis through natural language.
    52
    MIT

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/nayagamez/codex-cli-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server