What can you do with this server?

This server provides MCP tools to give text-only LLM agents vision and document understanding capabilities using Zhipu GLM-4.6V-Flash. It offers three tools: * Image Understanding (vision_analyze_image): analyze local images (auto base64) or public URLs. * Video Understanding (vision_analyze_video): analyze public video URLs (local must be uploaded first). * File Understanding (vision_analyze_file): analyze public document URLs (PDFs, text files; local must be uploaded first). All tools support an optional thinking mode for deeper inference. Responses are structured JSON with success fields (ok, content, thinking, usage) or error objects (ok:false, error with code and message). Error handling includes retries with exponential backoff (up to 3 times for 429/5xx/timeout), 60-second timeout, and categorized error codes (AUTH_ERROR, FILE_NOT_FOUND, TIMEOUT, API_ERROR, MODALITY_NOT_SUPPORTED).

How do I use deepseek-vision-mcp?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@deepseek-vision-mcp Analyze this image and describe what's in it: /tmp/screenshot.png" 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-vision-mcp

by JunHua-ECJTU

Overview Schema Related Servers Score Discussions

Python

Local

👁️ deepseek-vision-mcp

让纯文本大语言模型 Agent（如 DeepSeek 系）获得「眼睛」的 MCP 项目：通过标准 MCP 工具调用智谱 GLM-4.6V-Flash（免费视觉模型），完成图片 / 视频 / 文件三种模态的理解，并以纯文本结果回传给 Agent。

底层模型：智谱 GLM-4.6V-Flash（免费，128K 上下文，支持思考模式）官方文档：https://docs.bigmodel.cn/cn/guide/models/free/glm-4.6v-flash

特性

三模态理解：图片 / 视频 / 文件（PDF、文本等）一次提问，纯文本返回
本地图片免上传：传本地路径自动转 base64，无需公网 URL；视频/文件模态接受公网 URL
思考模式：可选开启（thinking=true），适合需要深度推理的视觉任务
健壮性：3 次指数退避重试（429 / 5xx / 超时）、60s 超时、结构化错误返回（绝不向 Agent 抛未处理异常）
标准 MCP（stdio）：兼容 Claude Code / Cursor / Cline / Continue / Windsurf / Claude Desktop / Reasonix 等任何 MCP 客户端
零成本：基于智谱免费模型，无需充值

架构设计

MCP 执行层 + Skill 决策层，职责分离：

┌─────────────────────────────────────────────────────┐
│  文本模型 Agent（DeepSeek 等）—— 只懂文字            │
│  · 读取 Skill 的业务规则，决定何时调用什么工具        │
└──────────────────────┬──────────────────────────────┘
                       │ MCP 协议（stdio）
┌──────────────────────▼──────────────────────────────┐
│  MCP Server（本仓库，Python）—— 真实执行             │
│  · 读本地文件 / 图片 → base64 编码                    │
│  · 调用智谱 API（认证、重试、超时、异常分类）          │
│  · 提供标准化工具：vision_analyze_image / _video / _file │
└──────────────────────┬──────────────────────────────┘
                       │ HTTPS
┌──────────────────────▼──────────────────────────────┐
│  智谱 GLM-4.6V-Flash API（视觉理解，返回文本）        │
└─────────────────────────────────────────────────────┘

层	职责	位置
MCP	真实执行：读文件、base64、调 API、重试、异常捕获、统一错误结构	`deepseek_vision_mcp/`
Skill	业务规则：何时调用、传什么参数、输出格式、降级策略	`skills/vision-agent/SKILL.md`

快速开始

① 安装

pip install "deepseek-vision-mcp @ https://github.com/JunHua-ECJTU/deepseek-vision-mcp/releases/latest/download/deepseek_vision_mcp-0.2.0-py3-none-any.whl"

② 配置 Key

# Windows PowerShell
$env:ZHIPU_API_KEY = "你的Key"
# macOS / Linux
export ZHIPU_API_KEY="你的Key"

Key 免费申请：登录智谱开放平台 → 个人中心 → API Keys。

③ 注册到 Agent（项目根目录创建 .mcp.json）

{
  "mcpServers": {
    "deepseek-vision": {
      "command": "python",
      "args": ["-m", "deepseek_vision_mcp.server"],
      "env": { "ZHIPU_API_KEY": "你的Key" }
    }
  }
}

④ 提问

分析这张图片：https://cdn.bigmodel.cn/static/logo/register.png

各 Agent 的详细注册方式与常见问题见下文安装与部署。

安装与部署

1. 前置条件

项目	要求
Python	≥ 3.10（Windows / macOS / Linux 均可）
智谱 API Key	免费申请：登录智谱开放平台 → 个人中心 → API Keys（GLM-4.6V-Flash 是免费模型，无需充值）

检查 Python 版本：

python --version    # 或 python3 --version

若未安装 Python：Windows 用 winget install Python.Python.3.12 --scope user，macOS 用 brew install python@3.12，Linux 用 apt install python3 等。

2. 安装软件包（四种方式任选其一）

方式 A：一条命令直接安装（推荐）

pip install "deepseek-vision-mcp @ https://github.com/JunHua-ECJTU/deepseek-vision-mcp/releases/latest/download/deepseek_vision_mcp-0.2.0-py3-none-any.whl"

升级版本后，请同步把 URL 中的 0.2.0 替换为新版本号。

国内网络下载依赖较慢时加镜像源（wheel 本体仍从 GitHub 拉取，只有依赖走镜像）：

pip install "deepseek-vision-mcp @ https://github.com/JunHua-ECJTU/deepseek-vision-mcp/releases/latest/download/deepseek_vision_mcp-0.2.0-py3-none-any.whl" -i https://pypi.tuna.tsinghua.edu.cn/simple

方式 B：手动下载 wheel 后安装

打开 Release 页面：https://github.com/JunHua-ECJTU/deepseek-vision-mcp/releases/tag/v0.2.0
在 Assets 区下载 deepseek_vision_mcp-0.2.0-py3-none-any.whl
本地安装：pip install deepseek_vision_mcp-0.2.0-py3-none-any.whl

方式 C：源码包（sdist，含 skills/ 决策层文件与测试）

下载 deepseek_vision_mcp-0.2.0.tar.gz 后：pip install deepseek_vision_mcp-0.2.0.tar.gz

方式 D：克隆仓库开发安装

git clone https://github.com/JunHua-ECJTU/deepseek-vision-mcp.git
cd deepseek-vision-mcp && pip install -e ".[dev]"

不想污染全局环境时，先建虚拟环境再安装：
python -m venv .venv
# Windows：.venv\Scripts\activate    macOS/Linux：source .venv/bin/activate
pip install ...（上面任一方式）

验证安装成功：

python -c "from deepseek_vision_mcp.server import main; print('deepseek-vision-mcp OK')"

3. 配置 API Key（三选一）

方式 1：环境变量（临时，当前终端有效）

# Windows PowerShell
$env:ZHIPU_API_KEY = "你的Key"
# macOS / Linux
export ZHIPU_API_KEY="你的Key"

方式 2：.env 文件（推荐；MCP server 启动时自动读取当前工作目录的 .env）

echo "ZHIPU_API_KEY=你的Key" > .env

方式 3：MCP 配置文件的 env 字段（见下文第 4 节各 Agent 模板）

4. 注册到你的 Agent（MCP 客户端）

通用 .mcp.json（Claude Code / Cursor / Cline / Continue / Windsurf 等）——项目根目录创建 .mcp.json：

{
  "mcpServers": {
    "deepseek-vision": {
      "command": "python",
      "args": ["-m", "deepseek_vision_mcp.server"],
      "env": { "ZHIPU_API_KEY": "你的Key" }
    }
  }
}

Claude Desktop——菜单 → Settings → Developer → Edit Config，写入同样的 JSON 后重启。

Reasonix——编辑 %APPDATA%\reasonix\config.toml（Windows）或 ~/.reasonix/config.toml：

[[plugins]]
name = "deepseek-vision"
command = "python"
args = ["-m", "deepseek_vision_mcp.server"]
env = { ZHIPU_API_KEY = "${ZHIPU_API_KEY}" }
startup_timeout_seconds = 60
call_timeout_seconds = 120

并把 Key 放入与 config.toml 同目录的 .env（${VAR} 由 Reasonix 从环境展开，密钥不入配置文件）；重启或点「刷新插件」。

其他 MCP 客户端——任何支持 stdio MCP 的客户端都等价于：

字段	值
`command`	`python`
`args`	`["-m", "deepseek_vision_mcp.server"]`
`env`	`ZHIPU_API_KEY=<你的Key>`

Windows 下 python 不在 PATH 时，把 command 换成完整路径，如 C:\Users\你的用户名\AppData\Local\Programs\Python\Python312\python.exe。

5.（可选）安装 Skill 业务规则层

Skill 指导 Agent「何时调用工具、传什么参数、输出什么格式、失败怎么降级」。从源码包（方式 C）或仓库中取出 skills/vision-agent/ 目录，复制到你的 Agent 的 skills 目录（各 Agent 约定不同，通常是 ~/.agent/skills/ 或项目 .agent/skills/），重启会话生效。

6. 验证是否可用

方法 1：MCP Inspector 官方工具

npx @modelcontextprotocol/inspector python -m deepseek_vision_mcp.server

应看到 3 个工具：vision_analyze_image / vision_analyze_video / vision_analyze_file。

方法 2：直接问你的 Agent

分析这张图片：https://cdn.bigmodel.cn/static/logo/register.png

或本地图片：

帮我看一下 C:\photo\receipt.jpg 里的金额是多少

7. 常见问题

现象	处理
工具返回 `AUTH_ERROR`	Key 未配置或无效，检查第 3 节
工具返回 `MODALITY_NOT_SUPPORTED`	视频/文件传了本地路径——视频/文件模态只接受公网 URL，先上传到可访问地址
工具返回 `API_ERROR`（HTTP 429）	智谱免费模型限流（"访问量过大"），稍后重试或避开高峰
启动报 `ModuleNotFoundError: mcp`	依赖未装全，补装：`pip install mcp httpx python-dotenv`
Agent 找不到工具	MCP 注册配置有误；检查 JSON 语法与 `command` 是否可执行

更多细节（各 Agent 配置、MCP Inspector 用法、故障排查表）见 docs/DEPLOY.md。

工具与使用示例

工具	说明	关键参数
`vision_analyze_image`	图片理解（本地路径或 URL）	`source`、`question`、`thinking`
`vision_analyze_video`	视频理解（仅公网 URL）	`source`、`question`、`thinking`
`vision_analyze_file`	文件理解（仅公网 URL，PDF/文本等）	`source`、`question`、`thinking`

示例：让 Agent「描述这张图片」后，工具返回统一 JSON 结构：

{
  "ok": true,
  "content": "图片中是一个深蓝色背景的 Logo，上面有白色文字……",
  "thinking": "",
  "usage": { "prompt_tokens": 123, "completion_tokens": 45 }
}

失败时返回 {"ok": false, "error": {"code": "...", "message": "..."}}，错误码包括 FILE_NOT_FOUND、UNSUPPORTED_FORMAT、AUTH_ERROR、API_ERROR、TIMEOUT、MODALITY_NOT_SUPPORTED、INTERNAL 等。

测试

pytest

注意事项

GLM-4.6V-Flash 不支持同时理解多种模态（图片/视频/文件一次只传一种）——Skill 层已约束
视频/文件模态要求可访问的公网 URL（本地文件需先上传到可访问位置）；仅图片支持本地路径
.env 已加入 .gitignore，API Key 永不入库
本包目前通过 GitHub Release 分发（未发布到 PyPI）

许可证

Available Tools

3 tools

vision_analyze_fileA

分析一个文档文件（PDF / 文本等）并回答问题。source 仅支持公网可访问的文件 URL；本地文件请先上传到可访问地址。question 为对文档的提问；thinking=True 开启深度思考。返回 JSON：{ok, content, thinking, usage} 或 {ok:false, error:{code,message}}。

ParametersJSON Schema

Name	Required	Description	Default
`source`	Yes
`question`	Yes
`thinking`	No

Output Schema

ParametersJSON Schema

Name	Required	Description
`result`	Yes

TDQS

A4.6/5.0

Behavior4/5

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

With no annotations, the description carries the burden of behavioral disclosure. It honestly reveals the input constraint (public URL only), the effect of thinking parameter, and the exact return JSON structure (both success and error). It does not mention potential rate limits or file size limits, but the provided details are substantial and useful.

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 three concise sentences, front-loaded with the primary purpose. Every sentence adds value: the first states what it does, the second covers constraints, and the third explains parameters and return format. No filler.

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

Completeness5/5

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

Given the tool's moderate complexity (3 parameters, no annotations, no output schema in structured form), the description covers all necessary aspects: input constraints, parameter meanings, return format, and error handling. It is sufficient for an agent to select and invoke the tool correctly.

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

Parameters5/5

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

Schema description coverage is 0%, but the description fully compensates by explaining each parameter: source (file URL, public only), question (question about the document), and thinking (enable deep thinking). This adds all necessary meaning beyond the bare schema definitions.

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 analyzes a document file (PDF/text) and answers questions, with a specific verb and resource. It distinguishes itself from sibling tools (vision_analyze_image, vision_analyze_video) by focusing on documents, making its purpose immediately clear.

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: source must be a publicly accessible URL, and local files need to be uploaded first. It also explains the thinking flag. However, it does not explicitly contrast with sibling tools, leaving the when-to-use decision implied by the file type rather than stated.

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

vision_analyze_imageA

分析一张图片并回答问题。source 为本地图片路径（自动转 base64）或公网 URL； question 为对图片的提问；thinking=True 开启深度思考模式（更慢但更深入）。返回 JSON：{ok, content, thinking, usage} 或 {ok:false, error:{code,message}}。

ParametersJSON Schema

Name	Required	Description	Default
`source`	Yes
`question`	Yes
`thinking`	No

Output Schema

ParametersJSON Schema

Name	Required	Description
`result`	Yes

TDQS

A4.8/5.0

Behavior5/5

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

No annotations are present, so the description carries the full burden. It discloses the auto-base64 conversion for local paths, the thinking mode's trade-off, and the exact JSON return structure including error format.

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 compact paragraph that front-loads the primary action, then explains parameters, and ends with return format. No wasted words.

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

Completeness5/5

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

Despite having an output schema, the description also specifies the return structure. It covers all parameters, input types, and the optional thinking mode. It is complete for a straightforward analysis tool.

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

Parameters5/5

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

Schema coverage is 0%, but the description explains all three parameters (source, question, thinking) in detail, including allowed values and effects. This fully compensates for the schema's lack of descriptions.

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 it analyzes an image and answers questions, using the verb '分析' (analyze) and resource '图片' (image). This differentiates it from sibling tools that analyze video or files.

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?

It explains that source can be a local path or URL, and question is the query, but it doesn't explicitly contrast with video/file tools. The scope is implied by the tool's name and the description.

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

vision_analyze_videoA

分析一段视频并回答问题。source 仅支持公网可访问的视频 URL（如 https://...）；本地视频请先上传到可访问地址。question 为对视频的提问；thinking=True 开启深度思考。返回 JSON：{ok, content, thinking, usage} 或 {ok:false, error:{code,message}}。

ParametersJSON Schema

Name	Required	Description	Default
`source`	Yes
`question`	Yes
`thinking`	No

Output Schema

ParametersJSON Schema

Name	Required	Description
`result`	Yes

TDQS

A4.6/5.0

Behavior4/5

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

With no annotations provided, the description carries the full burden of behavioral disclosure. It reveals the URL-only restriction, the effect of thinking=True, and the exact return JSON structure including both success and error formats. Though it omits details like rate limits or video size constraints, it covers the key behavioral aspects for this tool's scope.

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 neatly structured: a one-line purpose statement followed by specific input constraints, parameter definitions, and return format. Each sentence adds meaningful information without redundancy, making it concise and efficiently front-loaded.

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

Completeness5/5

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

For a tool with no annotations and a sparse schema, the description covers all necessary context: purpose, usage constraints, parameter semantics, and return format. The output schema exists but the description even summarizes the response shape, leaving no significant gaps for an agent to invoke the tool correctly.

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

Parameters5/5

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

The schema has zero description coverage, so the description fully compensates by explaining every parameter: source's URL requirement and handling of local files, question's role, and thinking's boolean toggle meaning. This exceeds what the schema provides and clarifies how to use each argument correctly.

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's function: 'Analyze a video and answer questions.' This uses a specific verb and resource, and naturally distinguishes it from sibling tools like vision_analyze_image and vision_analyze_file by focusing on video input.

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?

It provides explicit context on usage prerequisites: source must be a publicly accessible URL and local videos must be uploaded first. While it doesn't explicitly compare to alternatives or offer exclusion criteria, the video-specific purpose and constraints give clear operational guidance.

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

Install Server

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

A4.7/5.0

Disambiguation5/5

Each tool targets a distinct media type: image, video, and document file. The purpose of each is clearly separated by the input format, leaving no ambiguity about which tool to use for a given source.

Naming Consistency5/5

All tools follow the consistent pattern `vision_analyze_<type>`, making it easy to predict the tool name for new media types. The verb `analyze` and prefix `vision_` are used uniformly.

Tool Count5/5

Three tools cover the core capabilities of the server (image, video, and document analysis) without unnecessary bloat. This is a well-scoped set for a vision-focused server.

Completeness4/5

The set covers the primary media types (image, video, document), but local video and file inputs require public URLs, which could be a usability gap. Missing audio analysis is a minor omission but not core to vision.

Maintenance

ActivityMaintained

ResponsivenessSyncing

Resources

GitHub Repository

Need Help?

Related Servers

Unclaimed servers have limited discoverability.

Looking for Admin?

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

Related MCP Connectors

mcp-qwen-image
io.github.AceDataCloud
MCP server for Qwen Image 3 AI image generation
AI & Machine Learning Image & Video Processing
1
mcp-aichat
io.github.AceDataCloud
MCP server for AI dialogue using various LLM models via AceDataCloud
AI & Machine Learning
2
mcp-glm
io.github.AceDataCloud
MCP server for GLM chat completions using Zhipu AI models via AceDataCloud
AI & Machine Learning Developer Tools
2
Frenchie A
io.github.Lab94
OCR, transcription, file extraction, and image generation for AI agents via MCP.
AI & Machine Learning Image & Video Processing Speech Processing
72

Related MCP Servers

Luma MCP
AI & Machine Learning Image & Video Processing
JochenYang
A
license
A
quality
A
maintenance
Multi-model vision understanding MCP server that provides unified image analysis for AI assistants without native vision, supporting models like GLM-4.6V, DeepSeek-OCR, Qwen3-VL-Flash, and more.
Updated a month ago (2026-08-09 06:39 UTC)
1
947
113
MIT
glm-vision-mcp
Image & Video Processing AI & Machine Learning Multimedia Processing
qy564
F
license
Not graded
quality
C
maintenance
An MCP server that leverages Zhipu's free GLM-4.6V-Flash vision model to enable image, video, and file understanding (OCR, table parsing, defect detection, document Q&A, and more) across MCP-compatible clients like Codex and Claude Desktop.
Updated a month ago (2026-08-04 16:11 UTC)
glm-vision
Image & Video Processing AI & Machine Learning
yblt
F
license
Not graded
quality
C
maintenance
MCP server that gives text-only models vision capabilities via free GLM vision models, supporting image description, OCR, chart/document analysis, and grounding with automatic model fallback.
Updated a month ago (2026-08-04 08:11 UTC)
1
GLM-4.6V-Flash MCP Server
Image & Video Processing AI & Machine Learning
chiyan11
F
license
A
quality
B
maintenance
A Model Context Protocol server that wraps the free GLM-4.6V-Flash vision model, enabling text-only LLM clients like Codex, Cursor, and Claude Desktop to analyze images, videos, and files (PDF/TXT) through standard MCP tools.
Updated a month ago (2026-08-04 08:54 UTC)
3
2

Latest Blog Posts

Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)
By Om-Shree-0709 on July 25, 2026.
mcp
Agent Identity
OAuth 2.1
Your AI Chatbot Just Exposed Your CEO's Salary to an Intern
By Om-Shree-0709 on July 2, 2026.
Agent Identity
MCP Security
OAuth Delegation
Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)
By Om-Shree-0709 on June 30, 2026.
Agentic Ai
Prompt Injection
WebAssembly

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/JunHua-ECJTU/deepseek-vision-mcp'

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