deepseek-harness-codex-bridge
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@deepseek-harness-codex-bridgeDelegate the budget calculation bug to DeepSeek Harness and monitor progress"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
DeepSeek Harness ↔ Codex Bridge
An unofficial community project for DeepSeek Harness and Codex. It is not affiliated with or endorsed by DeepSeek or OpenAI.
A local, bidirectional collaboration bridge with no database or daemon:
Codex delegates bounded subtasks to DeepSeek Harness (DSH) through MCP.
DSH and Codex exchange persistent JSON mailbox messages.
While DSH is running, it can proactively create or continue a real Codex task through
codex app-server.A task ledger records ownership, status, workspace, and the linked Codex thread.
DSH cannot return the assigned subtask unchanged; non-blocker messages must include progress or evidence.
Codex remains the lead: it splits large work, coordinates dependencies, verifies the result, and makes the final decision. An assigned DSH subtask stays with DSH unless it reports material progress/evidence or a concrete blocker.
Requirements
macOS or Linux (Windows npm
.cmdshim behavior is not yet verified)Node.js
>=22.19.0DeepSeek Harness, installed and configured
Codex CLI with
codex app-serverfor proactive DSH → Codex chats
npm install --global @deepseek-ai/dsh
dsh webConfigure a model in the DSH Web UI before using the bridge.
Related MCP server: backchannel
Install
npm install --global https://github.com/Aloneswork/deepseek-harness-codex-bridge/releases/download/v0.2.2/deepseek-harness-mcp-bridge-0.2.2.tgz
codex mcp add deepseek_harness -- deepseek-harness-mcpThe package installs two commands:
deepseek-harness-mcp: stdio MCP server for Codex or another MCP client.dsh-codex-mail: mailbox and proactive-chat CLI available to DSH through Bash.
Recommended Codex MCP settings:
[mcp_servers.deepseek_harness]
tool_timeout_sec = 900
enabled_tools = [
"delegate_to_deepseek",
"list_deepseek_messages",
"reply_to_deepseek",
"get_bridge_task_status",
]
default_tools_approval_mode = "approve"Only use automatic MCP approval after the user has authorized calls without confirmation. This does not expand authorization for destructive, external, paid, publishing, credential, or sensitive-data actions.
The MCP client's tool-call timeout must be at least as long as DSH_BRIDGE_TIMEOUT_MS. A shorter client timeout can abandon an in-flight DSH process; the bridge now terminates tracked DSH children when its MCP process exits, but matching timeouts remains the clean path.
MCP tools
delegate_to_deepseek: creates a DSH-owned task and runs it headlessly.list_deepseek_messages: reads DSH progress, evidence, results, questions, and blockers.reply_to_deepseek: writes a persistent reply that DSH can read.get_bridge_task_status: reads task ownership, state, workspace, and linked Codex thread.
DSH mailbox CLI
DSH receives rootTaskId and taskId in every delegated prompt.
dsh-codex-mail send \
--root-task-id ROOT_ID \
--task-id TASK_ID \
--kind milestone \
--body "Parser inspection completed" \
--progress "100%" \
--evidence "fixture passed"
dsh-codex-mail inbox --task-id TASK_ID --mark-read
dsh-codex-mail status --task-id TASK_IDAllowed kinds are question, blocker, milestone, result, and reply. A blocker is itself an explicit blocking report; every other DSH message needs --progress or --evidence.
Proactive DSH → Codex chat
dsh-codex-mail chat \
--root-task-id ROOT_ID \
--task-id TASK_ID \
--kind question \
--body "Should the parser keep this fallback?" \
--progress "All fixtures pass" \
--evidence "test output attached" \
--workspace /absolute/project/pathThe first call creates and names a Codex task. Later calls for the same taskId resume its stored Codex thread. The Codex answer is both printed to DSH and stored in DSH's inbox.
This is active initiation, not a resident autonomous process: DSH can start a Codex task while DSH is executing. No agent can initiate after its process has stopped unless an external scheduler starts it again.
Storage and environment
Messages and task records default to ~/.dsh-codex-bridge/; message files are written atomically with local-only file permissions. No API keys are copied into the package or mailbox. Do not commit or share the mailbox because delegated prompts and agent replies may contain project data.
Optional environment variables:
DSH_BIN: DSH executable; defaultdshCODEX_BIN: Codex executable; defaultcodexDSH_CODEX_MAILBOX_DIR: shared mailbox directoryDSH_BRIDGE_TIMEOUT_MS: DSH timeout, default900000DSH_CODEX_CHAT_TIMEOUT_MS: proactive Codex turn timeout, default900000
Delegated task prompts are capped at 16,000 characters so they fit conservative process argument limits. Put large inputs in workspace files and delegate by path.
Collaboration policy
Recommended global ~/.codex/AGENTS.md policy:
- Codex is the primary agent and owns task framing, decomposition, coordination, decisions, integration, verification, and the final answer.
- Codex may call DSH without asking first when the user has authorized that workflow.
- DSH must execute an assigned subtask; it must not return or re-delegate the same subtask unchanged.
- DSH may contact Codex only with progress, evidence, a result, a progress-backed question, or an explicit blocker.
- Avoid simultaneous edits to the same files. DSH delegation never expands the user's authorization.
- After each DSH call, report what was delegated, what DSH returned or changed, and what Codex accepted, changed, or rejected.中文速用
安装并在
dsh web中配置 DeepSeek Harness。从 GitHub Release 全局安装本包:
npm install -g https://github.com/Aloneswork/deepseek-harness-codex-bridge/releases/download/v0.2.2/deepseek-harness-mcp-bridge-0.2.2.tgz。注册 MCP:
codex mcp add deepseek_harness -- deepseek-harness-mcp。Codex 用 MCP 委派与收发消息;DSH 用
dsh-codex-mail发信、收信、查状态或主动创建 Codex 任务。Codex 负责大任务拆解、协调、验收和最终决策;DSH 接到子任务后必须实做,不能原样推回。
The bridge is MIT licensed. See LICENSE.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- FlicenseBqualityDmaintenanceAn MCP server that exposes a library of delegation prompts to orchestrate tasks between a primary LLM and specialized sub-agents. It enables the execution of self-contained, bounded tasks with built-in support for configuration discovery and project-specific delegation libraries.3
- Alicense-qualityBmaintenanceMCP server for async messaging between AI coding agents, enabling cross-harness and cross-machine communication with Slack-like semantics and mail-shaped delivery.2MIT
- FlicenseBqualityCmaintenanceA Codex MCP server that delegates exploration and implementation tasks to DeepSeek via Claude Agent SDK, supporting subagent types and task persistence.21
- Alicense-qualityCmaintenanceA local MCP server that manages Claude-to-Codex task orchestration on Windows, providing tools for creating, dispatching, and monitoring tasks with safety constraints.MIT
Related MCP Connectors
A paid remote MCP for OpenAI Codex agent coordination MCP, built to return verdicts, receipts, usage
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Personal assistant MCP server with search, execute, packages, jobs, secrets, and integrations.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
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/Aloneswork/deepseek-harness-codex-bridge'
If you have feedback or need assistance with the MCP directory API, please join our Discord server