Reka Vision MCP Server

Reka Vision MCP Server

Enables AI agents to upload, index, search, and analyze videos through the Reka Vision API, supporting natural language search, visual question answering, and extraction of transcripts and captions.

Category
访问服务器

README

Reka Vision MCP Server

mcp-name: ai.reka/mcp

MCP server that lets AI agents upload, index, search, and analyze videos through the Reka Vision API. Agents can search across videos using natural language, ask questions about video content with visual analysis, detect objects in specific time ranges with text prompts, and read processed data like transcripts, captions, and scenes.

Quick Start

# Run the published local stdio server with your Reka API key
REKA_VISION_API_KEY="your-api-key" uvx reka-mcp

For MCP clients that need explicit command configuration:

{
  "command": "uvx",
  "args": ["reka-mcp"],
  "env": {
    "REKA_VISION_API_KEY": "your-api-key-here"
  }
}

For local development:

uv sync
uv run pre-commit install
REKA_VISION_API_KEY="test-key" uv run reka-mcp

The default mode is local stdio. In local mode, REKA_VISION_API_KEY is read once from the process environment and used for all requests.

Hosted Mode

Hosted mode runs the same MCP tools over Streamable HTTP. It does not use a process-wide Reka API key. Instead, each MCP HTTP request must include the user's key in X-Reka-API-Key; the server forwards that value to the Reka Vision API as x-api-key for that request only.

Production-style hosted startup:

REKA_MCP_MODE=hosted \
REKA_MCP_HTTP_HOST=0.0.0.0 \
PORT=8080 \
uv run reka-mcp

Endpoints:

  • MCP Streamable HTTP: http://<host>:<port>/mcp
  • Health check: http://<host>:<port>/health

Hosted clients must send:

X-Reka-API-Key: your-api-key

REKA_MCP_AUTH_TOKEN is optional MCP transport auth. When set, HTTP clients must also send Authorization: Bearer <token>.

index_video works the same in both modes: it polls the feature DAG until all requested features are ready (or times out). After upload_video, poll get_video until the video status is uploaded, then call index_video.

Run Hosted Mode Locally

Hosted mode enables DNS rebinding protection. Its default allowed hosts and origins are production/staging domains, so override them for localhost testing:

REKA_MCP_MODE=hosted \
REKA_MCP_TRANSPORT=http \
REKA_MCP_HTTP_HOST=0.0.0.0 \
REKA_MCP_HTTP_PORT=8080 \
PORT=8080 \
REKA_MCP_HTTP_PATH=/mcp \
REKA_MCP_ALLOWED_HOSTS="localhost:*,127.0.0.1:*" \
REKA_MCP_ALLOWED_ORIGINS="http://localhost:*,http://127.0.0.1:*" \
uv run reka-mcp

Then connect to http://localhost:8080/mcp and configure your MCP client or inspector to send X-Reka-API-Key. Check the server with:

curl -H "Host: localhost:8080" http://localhost:8080/health

Connect to Hosted MCP (mcp.reka.ai)

No installation required — connect directly to the hosted server with your Reka API key.

Claude Desktop

Claude Desktop does not speak Streamable HTTP directly, so use the mcp-remote bridge. Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "reka-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.reka.ai/mcp",
        "--header",
        "X-Reka-API-Key:${REKA_VISION_API_KEY}"
      ],
      "env": {
        "REKA_VISION_API_KEY": "your-api-key-here"
      }
    }
  }
}

Requires Node.js installed locally. Restart Claude Desktop after editing the config.

Claude Code

claude mcp add --transport http reka-mcp https://mcp.reka.ai/mcp \
  --header "X-Reka-API-Key: your-api-key-here"

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "reka-mcp": {
      "type": "streamable-http",
      "url": "https://mcp.reka.ai/mcp",
      "headers": {
        "X-Reka-API-Key": "your-api-key-here"
      }
    }
  }
}

Claude Desktop Setup (Local)

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "reka-mcp": {
      "command": "uvx",
      "args": ["reka-mcp"],
      "env": {
        "REKA_VISION_API_KEY": "your-api-key-here"
      }
    }
  }
}

Cursor Setup (Local)

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "reka-mcp": {
      "command": "uvx",
      "args": ["reka-mcp"],
      "env": {
        "REKA_VISION_API_KEY": "your-api-key-here"
      }
    }
  }
}

Claude Code Setup (Local)

claude mcp add reka-mcp -e REKA_VISION_API_KEY=your-api-key-here -- uvx reka-mcp

Updating

To update to the latest version, clear the cached package and restart your client:

uv cache clean reka-mcp

To check which version you're running:

uvx reka-mcp --version

Available Tools

Tool Description
upload_video Upload a video from a URL
list_videos List videos in your account or a group
get_video Get video details, metadata, and feature status
update_video Update a video's name, title, description, or group
delete_video Permanently delete a video and all indexed data
create_group Create a new video group
list_groups List all video groups
delete_group Delete a video group
index_video Index a video for search/QA/analysis. Waits until all requested features are ready (2-10 min).
search_videos Semantic search across indexed videos
ask_video Ask questions about video content with visual analysis
get_transcript Get transcript as text, segments, or words
get_captions Get AI-generated visual descriptions
get_scenes Get detected scene boundaries
get_feature_catalog List available features and dependencies
summarize_video Compact overview of video content and status

Environment Variables

Variable Default Description
REKA_MCP_MODE local Runtime mode: local or hosted
REKA_VISION_API_KEY (required in local mode) API key from https://platform.reka.ai. Not used as the primary auth source in hosted mode.
REKA_VISION_API_URL https://vision-agent.api.reka.ai API base URL
REKA_MCP_INDEX_TIMEOUT 600 Max seconds to wait for indexing
REKA_MCP_POLL_INTERVAL 5 Seconds between index status polls
REKA_MCP_TRANSPORT stdio in local, http in hosted Transport: stdio or http
REKA_MCP_HTTP_HOST 127.0.0.1 in local, 0.0.0.0 in hosted Host for HTTP transport
REKA_MCP_HTTP_PORT 8080 Port for HTTP transport. In hosted mode, PORT takes precedence when set.
REKA_MCP_HTTP_PATH /mcp Streamable HTTP endpoint path
REKA_MCP_ALLOWED_HOSTS mcp.reka.ai,staging.mcp.reka.ai in hosted Comma-separated allowed HTTP Host values for DNS rebinding protection
REKA_MCP_ALLOWED_ORIGINS https://mcp.reka.ai,https://staging.mcp.reka.ai in hosted Comma-separated allowed Origin values
REKA_MCP_AUTH_TOKEN (none) Optional bearer token for HTTP transport auth. Clients must send Authorization: Bearer <token> when set.

Release Checks

cd /path/to/reka-mcp
uv build
uv run twine check dist/*
# Publish only after explicit approval:
uv run twine upload dist/*

Example Workflows

Search and Visual Q&A

Agent: search_videos(query="revenue chart")
→ [{video_id: "v1", start: 30.0, end: 35.0, score: 0.95}]

Agent: ask_video(question="What numbers are on the chart?",
                 video_id="v1", start=30.0, end=35.0)
→ {answer: "Q3 revenue of $4.2M, up 32%...", conversation_id: "c1"}

Agent: ask_video(question="What's the percentage change?",
                 conversation_id="c1")
→ {answer: "Revenue increased by 32% quarter-over-quarter..."}

Cross-Video Comparison

Agent: search_videos(query="quarterly revenue")
→ [{video_id: "v1", start: 30.0, ...}, {video_id: "v2", start: 120.0, ...}]

Agent: ask_video(question="How do the revenue figures compare?",
                 videos=[
                   {video_id: "v1", start: 30.0, end: 35.0},
                   {video_id: "v2", start: 120.0, end: 125.0}
                 ])
→ {answer: "Video 1 shows Q3 at $4.2M while Video 2 shows Q4 at $5.1M..."}

Video Summary and Transcript Extraction

Agent: summarize_video(video_id="v1")
→ {name: "Lecture 3", duration_seconds: 3600, features: {...},
   scene_count: 42, transcript_preview: "Welcome to today's lecture..."}

Agent: get_transcript(video_id="v1", format="segments", start=0, end=60)
→ {data: [{start: 0.0, end: 5.2, text: "Welcome..."}, ...],
   total_count: 12, truncated: false}

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选
mcp-server-qdrant

mcp-server-qdrant

这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。

官方
精选
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选