Skip to main content
Glama
deenesjakoruzh
benzkittisak

codex-async-mcp

by benzkittisak
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

codex-async-mcp

codex CLIを非同期でラップするローカルMCPサーバー。ブロックせずにjob_idを即座に返すため、ClaudeがMCPプロトコルのタイムアウト（-32001）に達することはありません。

要件

  • Python 3.11以上

  • codex CLIがインストールされており、$PATHに含まれていること（v0.125.0以上）

  • Claude Code CLI

Related MCP server: codex-mcp-server

インストール

cd ~/payroll-mcp   # or wherever this repo lives
pip install -e ".[dev]"

確認:

python -c "from codex_async_mcp.server import mcp; print(mcp.name)"
# → codex-async-mcp

Claudeへの登録

グローバル（すべてのプロジェクト）

claude mcp add codex-async -s user -- python -m codex_async_mcp.server

プロジェクトのみ

cd ~/payrollservice-thailand   # or any project
claude mcp add codex-async -- python -m codex_async_mcp.server

確認

claude mcp list
# codex-async: python -m codex_async_mcp.server - ✓ Connected

ツール権限の追加 (settings.local.json)

{
  "permissions": {
    "allow": [
      "mcp__codex-async__codex_start",
      "mcp__codex-async__codex_poll",
      "mcp__codex-async__codex_list",
      "mcp__codex-async__codex_cancel"
    ]
  }
}

ツール

ツール

説明

codex_start(prompt, cwd, approval_policy?)

バックグラウンドでcodexを開始 → job_idを即座に返す

codex_poll(job_id, tail_lines?)

ステータスと出力の末尾を確認

codex_list(limit?)

最近のジョブを一覧表示（新しい順）

codex_cancel(job_id)

実行中のジョブを強制終了

approval_policyの値

値

Codexフラグ

動作

suggest

-s read-only

読み取り専用サンドボックス、書き込みなし

auto-edit

--full-auto

編集を自動適用

full-auto

--dangerously-bypass-approvals-and-sandbox

プロンプトなし、サンドボックスなし

Claudeの自動化には常にfull-autoを使用してください — suggestモードは対話的な入力を待ちますが、サブプロセス内では入力が届かないためです。

使用例

codex_start(
  prompt="In app/services/prorate_calculation_service.rb line 96, change format(...) to number_to_currency(...)",
  cwd="/Users/bbgummybear/payrollservice-thailand",
  approval_policy="full-auto"
)
# → { job_id: "f3a9b2", status: "running", pid: 12345 }

codex_poll(job_id="f3a9b2")
# → { status: "running", output: "Reading file..." }

codex_poll(job_id="f3a9b2")
# → { status: "done", exit_code: 0, output: "Applied changes to prorate_calculation_service.rb" }

ジョブの状態

ジョブは~/.codex-async/jobs/{job_id}/に保存されます:

~/.codex-async/jobs/f3a9b2/
  meta.json     ← status, pid, timestamps, exit_code
  output.txt    ← stdout + stderr from codex

meta.jsonの構造:

{
  "job_id": "f3a9b2",
  "status": "running | done | error | cancelled",
  "prompt": "...",
  "cwd": "/path/to/repo",
  "approval_policy": "full-auto",
  "pid": 12345,
  "started_at": "2026-04-29T10:00:00+00:00",
  "finished_at": null,
  "exit_code": null
}

トラブルシューティング

claude mcp listでcodex-async: ... - ✗ Failedと表示される

Pythonが見つからないか、パッケージが正しい環境にインストールされていません。

# Check which python Claude is using
which python

# If using conda, register with the full path
claude mcp add codex-async -s user -- /Users/bbgummybear/miniconda3/bin/python -m codex_async_mcp.server

# Verify the package is installed in that environment
/Users/bbgummybear/miniconda3/bin/python -c "import codex_async_mcp; print('ok')"

codex_startの直後にstatus: "error"になる

Codexの起動に失敗しました。生の出力を確認してください:

cat ~/.codex-async/jobs/<job_id>/output.txt

一般的な原因:

出力メッセージ

修正方法

command not found: codex

codexがPATHにありません — シェルプロファイルに追加するか、config.pyでCODEX_BINを設定してください

unknown flag: --dangerously-bypass-approvals-and-sandbox

Codexのバージョンが0.125.0未満です — npm install -g @openai/codexを実行してアップグレードしてください

permission denied

cwdが存在しないか、Claudeにアクセス権がありません

status: "running"のまま終了しない

サブプロセスがハングしています（入力を待機しているか、ループに陥っています）。

# Check if the process is still alive
ps aux | grep codex

# Check live output
tail -f ~/.codex-async/jobs/<job_id>/output.txt

# Cancel the job
codex_cancel(job_id="<job_id>")

最も一般的な原因: approval_policy="suggest"を使用しているため、対話的な承認を待機しています。代わりに"full-auto"を使用してください。

サーバー再起動後にジョブがstatus: "running"と表示される

MCPサーバーは再起動時にメモリ内のPopenレジストリを失います。次回のcodex_poll呼び出し時にPIDが終了していることが検出され、ステータスが自動的に更新されます。

codex_poll(job_id="<job_id>")
# → { status: "done", ... }   ← auto-resolved on first poll

古いジョブがディスクを圧迫している

# View all jobs sorted by date
ls -lt ~/.codex-async/jobs/

# Delete jobs older than 7 days
find ~/.codex-async/jobs -maxdepth 1 -type d -mtime +7 -exec rm -rf {} +

プロジェクト構造

codex-async-mcp/
├── README.md
├── pyproject.toml
├── src/
│   └── codex_async_mcp/
│       ├── __init__.py
│       ├── server.py        # MCP entry point, tool definitions
│       ├── job_manager.py   # spawn / poll / cancel / list
│       └── config.py        # JOBS_DIR, CODEX_BIN, defaults
└── tests/
    └── test_job_manager.py

テストの実行

pytest tests/ -v

Available Tools

4 tools
codex_cancelA

Cancel a running codex job by sending SIGTERM to the subprocess.

ParametersJSON Schema
NameRequiredDescriptionDefault
job_idYesThe job_id returned by codex_start.

Output Schema

ParametersJSON Schema
NameRequiredDescription

No output parameters

TDQS

A4.2/5.0
Behavior4/5

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

With no annotations, the description carries the full burden. It discloses that cancellation is done by sending SIGTERM to the subprocess, which is a key behavioral trait. It does not cover edge cases like job already finished, but the main behavior is well communicated.

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 a single 12-word sentence, front-loaded with the action, and contains no wasted words.

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

Completeness4/5

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

Given the tool's simplicity (one parameter, single action), the description is nearly complete. It could mention what happens if the job is not running, but the output schema likely handles error responses.

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?

The input schema already describes the only parameter (job_id) with 100% coverage. The description adds no new semantic information beyond what the schema 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 uses a specific verb ('Cancel') and resource ('running codex job'), and clearly distinguishes from siblings like codex_list, codex_poll, and codex_start.

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 implicitly states when to use (to cancel a running job with a job_id from codex_start) but does not explicitly state when not to use or provide alternative scenarios. However, the simplicity of the action mitigates the need for extensive guidelines.

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

codex_listA

List recent codex jobs with their status and prompt summaries.

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMax number of jobs to return (most recent first). Default: 20.

Output Schema

ParametersJSON Schema
NameRequiredDescription
resultYes

TDQS

A3.5/5.0
Behavior2/5

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

No annotations provided, so description carries full burden. It only states it lists jobs with status and summaries, but lacks details on pagination, ordering (though limit param says 'most recent first'), rate limits, or side effects. The description is too minimal for full transparency.

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?

Single sentence, no redundant words, front-loaded with the core action. Every word earns its place.

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?

Given the presence of an output schema (status and prompt summaries mentioned), the description is somewhat complete for a simple list operation. However, it lacks details on error handling, empty results, or additional behavioral context that would fully inform an agent.

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 coverage is 100% for the single parameter 'limit', and the schema itself provides a description including default and ordering. The tool description adds no extra meaning beyond what the schema already conveys.

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?

Clearly states the tool lists recent codex jobs, including status and prompt summaries. The verb 'list' and resource 'recent codex jobs' are specific and distinguish from sibling tools (cancel, poll, start) which are different actions.

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

Usage Guidelines3/5

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

No explicit guidance on when to use or avoid this tool. The purpose is implied by the name and description, but no alternatives or exclusions are mentioned. Siblings have distinct purposes, so usage is inferred but not clarified.

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

codex_pollA

Poll the status and output of a running (or finished) codex job.

ParametersJSON Schema
NameRequiredDescriptionDefault
job_idYesThe job_id returned by codex_start.
tail_linesNoHow many trailing lines of output to return. Default: 100.

Output Schema

ParametersJSON Schema
NameRequiredDescription

No output parameters

TDQS

A4/5.0
Behavior3/5

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

The description states the purpose but lacks details on behavioral traits such as whether the tool is idempotent or safe to call repeatedly. Since annotations are absent, the description carries the burden, and it only provides minimal transparency.

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 a single, front-loaded sentence with no extraneous words. It efficiently conveys the tool's purpose.

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

Completeness4/5

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

Given the presence of an output schema, the description does not need to explain return values. It covers the essential purpose and scope, though it could mention that the tool can be called multiple times safely.

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 coverage is 100%, and the description adds no additional meaning beyond what the schema already provides for the two parameters. The description's mention of 'output' hints at tail_lines, but this is redundant with the schema.

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 verb 'Poll' and the resource 'status and output of a running (or finished) codex job', which is specific and distinguishes it from sibling tools like codex_start, codex_cancel, and codex_list.

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 implies usage after a job is started, but does not explicitly provide when-not-to-use or alternatives. The context is clear enough for an agent to infer appropriate usage.

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

codex_startA

Start a codex task asynchronously in the background.

Returns a job_id immediately — does not block or timeout. Use codex_poll(job_id) to check progress.

ParametersJSON Schema
NameRequiredDescriptionDefault
promptYesThe task description to pass to codex.
cwdYesAbsolute path to the working directory for codex.
approval_policyNoOne of 'suggest', 'auto-edit', 'full-auto'. Default: 'suggest'.suggest

Output Schema

ParametersJSON Schema
NameRequiredDescription

No output parameters

TDQS

A4.2/5.0
Behavior4/5

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

No annotations provided, so description carries full burden. Discloses async, non-blocking, immediate return of job_id. Does not mention side effects or auth, but core behavior is adequately covered.

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?

Two sentences, no wasted words. Front-loaded with purpose and key behavior. Highly efficient.

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

Completeness4/5

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

Has output schema. Describes async nature and returns job_id. Could mention cancellation via sibling codex_cancel, but sufficient for a simple start tool.

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 baseline 3. Description does not add meaning beyond schema; each parameter is defined in schema. No extra context provided.

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?

Clearly states the verb 'Start', resource 'codex task', and key behavior 'asynchronously in the background'. Distinguishes from siblings by mentioning that codex_poll is used to check progress.

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?

Provides explicit guidance to use codex_poll for progress checking. Implicitly tells when to use this tool (async tasks) but lacks explicit when-not-to-use or alternatives beyond polling.

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.

  1. 4 tool updatesv0.1.0
    • First observedcodex_cancel
    • First observedcodex_list
    • First observedcodex_poll
    • First observedcodex_start
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.2/5.0

Scored across 4 tools

Disambiguation5/5

Each tool has a distinct purpose: start, list, poll, and cancel. There is no overlap in functionality, and the descriptions clearly differentiate them.

Naming Consistency5/5

All tools follow a consistent 'codex_verb' pattern using snake_case, making it predictable for an agent to infer tool behavior from the name.

Tool Count5/5

Four tools cover the essential operations for managing async jobs (start, list, poll, cancel) without redundancy or missing critical actions.

Completeness5/5

The tool set covers the full lifecycle of an async job: initiating (start), monitoring (poll, list), and termination (cancel). No obvious gaps are present.

Maintenance

ActivityInactive
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • codex-mcp-server

    Pigowenhsiao
    A
    license
    Not graded
    quality
    D
    maintenance
    Wraps OpenAI Codex CLI as an MCP server, exposing 8 Codex tools (exec, review, skill list, skill run, status, poll, list jobs, kill) as named tools for use with pi or codex.
    949 npm
    ISC

  • engineering-bridge

    wudy29
    A
    license
    Not graded
    quality
    A
    maintenance
    A local STDIO MCP server that bridges MCP clients to the Codex CLI by sending instructions to a configured workspace, exposing task run, status, and result tools with a read-only sandbox and no remote transport.
    133
    MIT