Pexels

io.github.developer-ishan由 GitHub 账号 developer-ishan 发布v1.0.2更新于 Oct 24, 2026

MCP server for searching Pexels stock photos, videos, and curated collections.

概览

AI 生成的概览

让助手搜索 Pexels 的免费图库照片、视频和精选合集,并把免版税素材引入工作流程。

可以做什么
把 Pexels API 封装为九个工具:照片搜索、精选照片流和按编号获取单张照片;视频搜索、热门视频和按编号获取单个视频;以及精选合集、已认证用户自己的合集和合集内的素材。返回结果是 Pexels 原始 JSON,并附加 _rateLimit 字段,包含 limit、remaining 和 reset,便于助手自行限流。结果中带有摄影师或摄像师署名。
适用场景
适合助手需要真实图库图片或视频的场景:为幻灯片和文档配图、为博客或社交媒体推荐缩略图、制作情绪板、寻找背景视频片段,或为编码任务生成占位图片。如果只需要照片且已接入通用图片服务器,则不必添加。
运行要求
通过 npx(npm 包 mcp-pexels)以 stdio 方式在本地运行,需要 Node.js 20 或更高版本。需要免费的 Pexels API 密钥,通过环境变量 PEXELS_API_KEY 提供,并需要访问 Pexels HTTPS API 的网络连接。仅支持桌面版 MCP 客户端;网页版不支持 stdio 服务器。
安装前请注意
PEXELS_API_KEY 密钥从本地进程环境读取,并通过 Authorization 请求头发送给 Pexels API;不要把它写进共享配置文件。该服务器对 Pexels 只做读取,但会把搜索词发送给第三方。免费额度有限(每小时 200 次、每月 20,000 次),且使用 Pexels 内容需署名平台,并尽可能署名摄影师。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Pexels,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

mcp-pexels — Pexels MCP server for Claude, Cursor, and other AI agents

[npm version] [npm downloads] [CI] [license: MIT] [Node ≥20] [MCP compatible]

A Model Context Protocol server that exposes the Pexels API — free stock photos, videos, and curated collections — as MCP tools that any compatible AI client (Claude Desktop, Claude Code, Cursor, Cline, Continue, etc.) can call directly from a chat.

📦 npm: mcp-pexels · 🏷 Latest release: v1.0.0 · 💻 Source: developer-ishan/mcp-pexels

Photos & videos provided by Pexels.

Drop-in config

json
{  "mcpServers": {    "pexels": {      "command": "npx",      "args": ["-y", "mcp-pexels"],      "env": { "PEXELS_API_KEY": "your_key_here" }    }  }}

Paste into your MCP client's config, restart, and the nine pexels_* tools are live. Get a free key at https://www.pexels.com/api/.


Table of contents


What can you build with this?

Once mcp-pexels is connected to an AI client, the agent can pull royalty-free photos and videos straight into its workflow. A few concrete uses:

  • Slide-deck and document illustration — "Find me a landscape photo of mountains at sunset for slide 3" → the agent searches Pexels, picks one, and embeds it.
  • Blog post / social media drafting — "Suggest three thumbnail options for an article about remote work" → returns three Pexels URLs with photographer credit.
  • Background research for video editing — "Show me popular drone footage under 30 seconds in 4K" → uses pexels_popular_videos with size and duration filters.
  • Mood boards — "Build me a mood board of warm-toned coffee shop photos" → searches with color: "orange".
  • Coding agents — generate CSS/HTML with realistic placeholder imagery instead of lorem-pixel boxes.
  • Curated collections — pull a specific Pexels collection (e.g. company-owned shots tagged on Pexels) into the agent's context.

Because every response includes a _rateLimit object, the agent can self-throttle and respect the Pexels free-tier quota (200 req/hour, 20 000/month).


Capabilities — nine tools

DomainToolWhat it does
Photospexels_search_photosSearch photos by query with orientation / size / color / locale filters.
Photospexels_curated_photosFetch the Pexels team's real-time curated photo feed.
Photospexels_get_photoRetrieve a single photo by its numeric id.
Videospexels_search_videosSearch videos by query with orientation / size / locale filters.
Videospexels_popular_videosFetch popular videos with optional width / height / duration constraints.
Videospexels_get_videoRetrieve a single video by its numeric id.
Collectionspexels_featured_collectionsList Pexels featured collections.
Collectionspexels_my_collectionsList the authenticated user's own collections.
Collectionspexels_collection_mediaRetrieve the photos and videos inside a specific collection.

Each tool response is the raw Pexels JSON merged with a top-level _rateLimit field exposing limit, remaining, and reset (UNIX timestamp).

See docs/tools/ for per-tool parameter tables and example payloads.


Connect to your MCP client

All paths below use npx -y mcp-pexels — no global install required. Replace your_key_here with a real Pexels API key.

Claude Desktop

Option 1 — one-click .mcpb install (recommended). Download the latest mcp-pexels-*.mcpb from the Releases page, double-click the file, paste your Pexels API key when Claude Desktop prompts, and click Install.

Option 2 — manual JSON config. Edit claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • Linux: ~/.config/Claude/claude_desktop_config.json
json
{  "mcpServers": {    "pexels": {      "command": "npx",      "args": ["-y", "mcp-pexels"],      "env": { "PEXELS_API_KEY": "your_key_here" }    }  }}

Restart Claude Desktop.

Claude Code

bash
claude mcp add pexels -e PEXELS_API_KEY=your_key_here -- npx -y mcp-pexels

Or edit ~/.claude.json directly and add the same mcpServers block as above.

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

json
{  "mcpServers": {    "pexels": {      "command": "npx",      "args": ["-y", "mcp-pexels"],      "env": { "PEXELS_API_KEY": "your_key_here" }    }  }}

Cline (VS Code)

Open the Cline extension settings → MCP Servers → Edit Settings. Add:

json
{  "mcpServers": {    "pexels": {      "command": "npx",      "args": ["-y", "mcp-pexels"],      "env": { "PEXELS_API_KEY": "your_key_here" }    }  }}

Continue.dev

In ~/.continue/config.json:

json
{  "experimental": {    "modelContextProtocolServers": [      {        "transport": {          "type": "stdio",          "command": "npx",          "args": ["-y", "mcp-pexels"],          "env": { "PEXELS_API_KEY": "your_key_here" }        }      }    ]  }}

Generic / other clients

If your client supports stdio MCP servers it just needs command, args, and env. Use the same shape as Cursor / Claude Desktop above. For inspector / debugging:

bash
export PEXELS_API_KEY=your_key_herenpx -y @modelcontextprotocol/inspector npx -y mcp-pexels

This opens a browser UI where you can call each tool interactively.


Example prompts

Once connected, try prompts like:

  • "Find me five landscape photos of mountains at golden hour. Make sure they're at least full HD."
  • "Get me popular Pexels videos shorter than 15 seconds, 1080p or higher, of city traffic."
  • "Pull photo id 2014422 and tell me the photographer."
  • "List the featured Pexels collections — I want one I can browse for cooking photos."
  • "Search videos of 'forest' in portrait orientation; pick the highest-quality HD file for me."
  • "Show me Pexels photos with predominantly orange tones suitable for a fall-themed newsletter header."

The agent will pick the appropriate tool, send the right parameters, and include photographer/videographer credit as part of its answer.


How it compares

mcp-pexelsRaw Pexels REST APIImage-only general MCP servers
Photos✅✅varies
Videos✅✅usually ❌
Collections✅✅❌
Rate-limit headers exposed to agent✅❌ (raw headers, agent must parse)❌
Zod-validated input✅❌varies
npx-installable, zero config✅❌varies
Type-safe TS source✅n/avaries
MIT-licensed✅n/avaries

If you only need photos and you already have a generic image MCP server wired up, that may be enough. If you want first-class video and collection support with rate-limit awareness, this is purpose-built for it.


FAQ

How do I get a Pexels API key?

Sign up at https://www.pexels.com/api/ — keys are issued instantly and the free tier allows 200 requests/hour and 20 000 requests/month.

Does this cost anything?

No. The Pexels API is free, the npm package is free (MIT), and mcp-pexels itself does not call any paid service.

Will this work offline?

No — the server is a thin wrapper over the Pexels HTTPS API and requires network access.

Can I use Pexels content commercially?

Yes, per the Pexels license, but you must attribute the platform and (when possible) the photographer. See Attribution.

Does it work with Claude.ai (browser)?

Claude.ai's web product doesn't yet support stdio MCP servers — only Claude Desktop, Claude Code, and the API do. Anthropic is rolling out browser MCP support, so this should change.

Is there a hosted / SaaS version?

No. Run it locally via npx. Because Pexels API keys are per-user, hosting a shared instance doesn't make sense.

How do I update to a newer version?

npx always fetches the latest published version. If you've installed globally, run npm update -g mcp-pexels.

Can I add more Pexels endpoints?

The Pexels API surface is fully implemented (9/9 endpoints as of Pexels API v1). New endpoints will be added via PR — see Contributing.

Does it work with @modelcontextprotocol/inspector?

Yes:

bash
export PEXELS_API_KEY=your_key_herenpx -y @modelcontextprotocol/inspector npx -y mcp-pexels

How is rate limiting handled?

Every tool response includes a _rateLimit object with limit, remaining, and reset (UNIX timestamp). The agent can read this and self-throttle. If Pexels returns 429, the server surfaces it as a PexelsApiError so the agent sees the failure.

Where does my API key go?

It is read from the PEXELS_API_KEY env var on the local process. It never leaves your machine except in Authorization headers to api.pexels.com. Nothing is logged.


Requirements

  • Node.js ≥ 20 (uses native fetch and --env-file)
  • A free Pexels API key — request one here

Development

bash
git clone https://github.com/developer-ishan/mcp-pexels.gitcd mcp-pexelsnpm installecho "PEXELS_API_KEY=your_key_here" > .env
npm test           # 42 mocked vitest cases, no networknpm run build      # tsc → dist/npm run dev        # tsx watchnpm run test:watch # vitest in watch modenpm start          # node --env-file=.env dist/index.js

Project layout

src/├── index.ts                  # MCP server entrypoint (stdio)├── http/│   ├── client.ts             # PexelsClient (injectable fetch for tests)│   └── types.ts              # PexelsApiError, RateLimit, PexelsResponse├── tools/│   ├── index.ts              # composes the three domain factories│   ├── photos.ts             # 3 photo tools│   ├── videos.ts             # 3 video tools│   └── collections.ts        # 3 collection tools└── types/    ├── tool.ts               # defineTool helper + ToolDefinition    ├── photos.ts    ├── videos.ts    └── collections.ts
tests/├── helpers/mockClient.ts     # makeMockClient() — fake fetch via vitest mocks├── photos.test.ts            # 17 cases├── videos.test.ts            # 13 cases└── collections.test.ts       # 12 cases

Testing approach

Tests construct a PexelsClient with a vi.fn() fetchImpl returned by makeMockClient(responses). No real HTTP is performed. Each tool is exercised for:

  1. Happy path — URL serialization and Authorization header.
  2. Error mapping — non-2xx → PexelsApiError with status and body.
  3. Zod validation — required params, enum membership, integer bounds.

Releasing

This repo uses a tag-driven release workflow (.github/workflows/release.yml):

  1. Bump the version: npm version patch (or minor / major) — creates a commit and tag.
  2. Push: git push && git push --tags.
  3. The workflow runs tests, builds, publishes to npm with provenance using the NPM_TOKEN repository secret, builds the .mcpb bundle via npm run pack:mcpb, and creates a GitHub release with auto-generated notes that includes the bundle as a downloadable asset.

See CHANGELOG.md for version history.


Contributing

Issues and pull requests welcome. See CONTRIBUTING.md for the development workflow and SECURITY.md to report a vulnerability privately.

If you ship mcp-pexels inside a tool or product, open a PR adding it to the "Used by" list in CONTRIBUTING.md.

Listed on / submit to

See docs/MARKETPLACES.md for the submission cheatsheet.


Attribution & License

When you use Pexels content, please follow the Pexels API guidelines:

  • Show a prominent link to Pexels — e.g. "Photos provided by Pexels".
  • Credit photographers/videographers when possible — e.g. "Photo by John Doe on Pexels".

The mcp-pexels package itself is released under the MIT License.

来源:README.md,提交 218c9a8

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v1.0.2最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分