codex-claude-agent-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ALLOWED_MODELS | No | Comma-separated allowed model overrides. | |
| CLAUDE_CLI_PATH | No | Path to the Claude Code CLI executable. | (bundled) |
| MAX_TASK_TIMEOUT_SEC | No | Hard ceiling for timeout_sec. | 7200 |
| ALLOWED_PROJECT_ROOTS | No | Optional ;-separated (Windows) / :-separated (POSIX) long-term cwd roots. Empty grants nothing; per-job project_root authorizes the rest. | |
| CLAUDE_MAX_CONCURRENCY | No | Max concurrent Claude sessions per MCP process. | 4 |
| DEFAULT_TASK_TIMEOUT_SEC | No | Default per-task timeout. | 3600 |
| CLAUDE_SDK_MAX_BUFFER_SIZE | No | Max bytes buffered per streamed SDK JSON line (default SDK limit is 1 MiB, which large tool_result payloads can exceed). | 20971520 |
| CODEX_CLAUDE_AGENT_MCP_DB_PATH | No | SQLite state path (shared across processes). | ~/.codex-claude-agent-mcp/state.db |
| CODEX_CLAUDE_AGENT_MCP_LOG_LEVEL | No | stderr log level. | INFO |
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 | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| pingB | Return pid/version, SDK/CLI metadata, buffer size; no CC call. |
| execute_taskC | Run one CC task without review; only a confirmed session is resumable. |
| review_taskC | Fresh Read/Grep/Glob-only CC review; acceptance is required; never repairs. |
| continue_taskC | Resume this job's CC-1 session with feedback; inherits the job's root. |
| run_jobD | Execute; if review=true and acceptance is nonempty, add a fresh review. |
| run_jobsC | Run an already-independent job batch with bounded concurrency. |
| get_job_statusC | Read persisted job state (sessions, stage statuses); no CC call. |
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 7 tools
execute_task and run_job both serve as execution entry points with overlapping purposes, and run_job can also incorporate review, blurring boundaries with review_task. Descriptions hint at differences (review option, scope) but the job vs task distinction is not defined, leaving room for misselection.
Mostly consistent snake_case verb_noun (execute_task, review_task, continue_task, run_job, run_jobs, get_job_status), but ping breaks the pattern and run/execute verbs are used interchangeably for similar actions. Still readable and predictable overall.
7 tools is well-scoped for an orchestration server covering health check, single/batch execution, review, session continuation, and status. Each tool has a distinct role without excessive surface area.
Core lifecycle is covered: start tasks/jobs, review, continue sessions, check status, and batch runs. However, missing cancellation/abort and job listing operations are notable gaps that could hinder cleanup or discovery in longer workflows.