Skip to main content
Glama
deenesjakoruzh
kira4094

DeepSeek Subagent MCP Server

by kira4094
OverviewSchemaSourceRelated ServersDiscussionsAdmin
JavaScript
Remote

DeepSeek Subagent MCP Server

Codex(또는 모든 MCP 클라이언트)가 하위 작업을 DeepSeek에 하위 에이전트로 위임할 수 있게 해주는 MCP 서버입니다. 단일 도구 deepseek_task에 작업(+ 선택적 컨텍스트)을 전달하면 DeepSeek의 최종 답변을 돌려받습니다.

기능

  • 🧠 단일 도구 deepseek_task — 독립적인 작업 위임(리서치/초안/검토/리팩터링/분석)

  • 💰 기본값은 deepseek-v4-flash(비용 효율적), deepseek-v4-pro는 환경 변수를 통해 전환

  • 🔍 reasoning_content(V4 사고 모드)를 <thinking> 블록으로 표시

  • ⚡ MCP 외 의존성 없음, npx 한 줄 배포

Related MCP server: deepseek-subagent-mcp

요구 사항

설치 및 실행

cd deepseek-subagent-mcp-server
npm install
npm start

환경 변수

변수

필수

기본값

설명

DEEPSEEK_API_KEY

✅

—

DeepSeek API 키입니다.

DEEPSEEK_MODEL

deepseek-v4-flash

모델. deepseek-v4-pro도 사용할 수 있습니다.

DEEPSEEK_BASE_URL

https://api.deepseek.com

엔드포인트 재정의(프록시용).

DEEPSEEK_MAX_TOKENS

8192

기본 최대 출력 토큰 수입니다.

Codex 설정

Codex의 MCP 구성에 추가하세요:

{
  "mcpServers": {
    "deepseek-subagent": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "deepseek-subagent-mcp-server"],
      "env": {
        "DEEPSEEK_API_KEY": "your-deepseek-api-key",
        "DEEPSEEK_MODEL": "deepseek-v4-flash"
      }
    }
  }
}

또는 로컬 체크아웃에서 실행하세요:

{
  "mcpServers": {
    "deepseek-subagent": {
      "command": "node",
      "args": ["D:\\GitHub\\MCP\\deepseek-subagent-mcp-server\\src\\index.js"],
      "env": {
        "DEEPSEEK_API_KEY": "your-key-here"
      }
    }
  }
}

도구: deepseek_task

파라미터

타입

필수

설명

task

string

✅

DeepSeek에 전달할 하위 작업입니다. 수행할 일과 출력 형식을 구체적으로 지정하세요.

context

string

선택적 컨텍스트 자료(코드, 메모, 스니펫).

system

string

선택적 사용자 지정 시스템 프롬프트(기본값: 하위 에이전트 페르소나).

max_tokens

number

최대 출력 토큰 수. 기본값 8192.

temperature

number

0-2. 생략하면 모델 기본값을 사용합니다.

사용 예시

# Delegate a code review
deepseek_task(task="Review this function for bugs and edge cases", context="<code>...")

# Delegate a research task
deepseek_task(task="Summarize the key risks of a subscription pricing model")

관련 프로젝트

라이선스

MIT

Available Tools

1 tool
deepseek_taskA

Delegate a sub-task to DeepSeek as a sub-agent and get back its result. Use this for work that can be done independently: research, drafting, code review, refactoring, analysis, writing, or any self-contained task that shouldn't pollute the main conversation. The caller (Codex) supplies a task description and optional context; DeepSeek completes the task in one shot and returns the final answer (with reasoning, if produced). Default model: deepseek-v4-flash. Override via DEEPSEEK_MODEL env (e.g. DeepSeek-V4-Pro-0813).

ParametersJSON Schema
NameRequiredDescriptionDefault
taskYesThe sub-task for DeepSeek to complete. Be specific: what to do, what format the result should take.
systemNoOptional custom system prompt overriding the default subagent persona.
contextNoOptional context material the sub-agent should consider (code snippets, file contents, notes).
max_tokensNoMaximum output tokens.
temperatureNoSampling temperature (0-2). Omit to use the model default.
reasoning_effortNoThinking effort for DeepSeek V4 reasoning mode. Omit to use the model default.

TDQS

A4.1/5.0
Behavior4/5

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

With no annotations, the description carries the burden of explaining behavior. It does this well: it clarifies the one-shot execution model, the return style ('returns the final answer (with reasoning, if produced)'), and the default/environment-configurable model. It doesn't cover failure modes or side effects, but the behavior of a delegation tool is well-scoped.

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

Conciseness4/5

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

The description is modestly long but every sentence carries weight: the first defines the operation and result, the second clarifies the use cases and scope, and the third provides model configuration with an override example. It is structured and front-loaded, without filler.

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?

The tool is a one-shot delegation with no output schema and no siblings. The description adequately conveys its oracle-style behavior, expected result, and the extent of inputs. It could give example output shape or failure handling, but not that could be unique; the essentials are present.

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%, so all parameters are already documented. The description restates some parameter-level guidance ('supplies a task description and optional context') but adds no new meaning beyond the schema. The mention of the env-var model override is useful but relates to configuration rather than the input parameters. Baseline 3 is appropriate.

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 ('Delegate'), a clear resource ('sub-task to DeepSeek as a sub-agent'), and explicitly frames the return value ('get back its result'). It further distinguishes the tool by noting the work is done independently as a sub-agent without polluting the main conversation — making the purpose unambiguous.

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 a clear list of when the tool is appropriate (research, drafting, code review, refactoring, analysis, writing, any self-contained task) and states that it is for independent work. It does not explicitly list exclusions or alternatives, but there are no sibling tools to differentiate from, so the guidance is contextually sufficient.

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. 1 tool updatev0.2.0
    • First observeddeepseek_task
Deploy Server
Context Every Agent Can ShareAgents share what they learn, hand off work, and keep it when you switch models. Start with 5 GB free.fulcradynamics.comAd

TDQS

A4.2/5.0

Scored across 1 tool

Disambiguation5/5

There is only one tool, so there is no possibility of confusion or overlapping purposes. The tool clearly covers the single action of delegating a sub-task to DeepSeek.

Naming Consistency5/5

With a single tool, there is no pattern to be inconsistent about. The name 'deepseek_task' is descriptive and follows a clear noun-based naming convention.

Tool Count3/5

One tool is on the thin side; while the server's purpose is narrow, a richer integration might include multiple tools (e.g., different task types or configuration options). The count is borderline but not unreasonable for a single-purpose sub-agent server.

Completeness4/5

The tool fully covers the stated purpose of delegating a self-contained task and returning the result. Minor gaps exist, such as no way to manage task history or configure multiple output modes, but these are not essential to the core workflow.

Maintenance

ActivitySlowing
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • deepseek-mcp

    arizen-dev
    A
    license
    A
    quality
    D
    maintenance
    Enables using DeepSeek models as a small, cheap supervised worker from any MCP-compatible client, providing fast flash and deep reasoning tools for bounded tasks.
    2
    53
    MIT

  • deepseek-subagent-mcp

    gaztrabisme
    A
    license
    A
    quality
    B
    maintenance
    Enables AI coding agents like Claude Code or Codex to delegate tasks to a DeepSeek Harness subagent with its own context window, providing tools for task delegation, result waiting, continuation, and supervision with sandboxed execution.
    6
    MIT