upwork-mcp

upwork-mcp

A Model Context Protocol server that enables AI agents to search, analyze, and apply for Upwork jobs autonomously. It facilitates specialized job scoring, proposal submission, and integrated client messaging through a combination of browser automation and API access.

Category
访问服务器

README

upwork-mcp

A Model Context Protocol (MCP) server that connects Claude to Upwork — enabling AI agents to search jobs, analyze opportunities, submit proposals, and manage client communications autonomously.

Works for any freelance niche: development, design, writing, marketing, automation, and more.


Features

Tool Description
search_jobs Search Upwork jobs by keyword, type, budget, experience level
get_job_details Fetch full job description, screening questions, and client history
analyze_job Score a job 0–100 across niche fit, client quality, budget, competition, and clarity
submit_proposal Automatically write and submit a proposal with custom cover letter and bid
get_proposals List submitted proposals and their current status
get_messages Read conversations and check for unread messages
send_message Reply to clients in ongoing conversations
get_profile View your freelancer profile, JSS score, and connects balance
update_profile Update your freelancer profile title, overview, and skills
manual_login Capture session cookies from an already-logged-in Chrome tab

Architecture

Claude Agent
    │
    ▼
MCP Gateway (stdio, never restarts)
    │  HTTP POST /tool
    ▼
Worker Server (port 47821, hot-reloads on code changes)
    │
    ▼
CDP Proxy (port 9223, rewrites Host headers)
    │
    ▼
Chrome on host (port 9222, Playwright CDP connection)
    │
    ▼
Upwork Web UI

The gateway/worker split means you can update tool logic without restarting Claude or your MCP client.


Prerequisites

  • Node.js 20+
  • Docker + Docker Compose
  • Google Chrome installed
  • An Upwork freelancer account

Quick Start (Docker)

1. Clone and configure

git clone https://github.com/zcrossoverz/upwork-mcp.git
cd upwork-mcp
cp .env.example .env

Edit .env:

UPWORK_EMAIL=your@email.com
UPWORK_PASSWORD=yourpassword

FREELANCER_NAME=Your Name
FREELANCER_TITLE=Your Professional Title
FREELANCER_NICHE=your,skills,here

BID_RATE_DEFAULT=40
BID_RATE_MIN=25
BID_RATE_MAX=100

2. Start Chrome with CDP

connect-chrome-docker.bat

This launches Chrome with remote debugging + a CDP proxy that allows Docker to connect.

3. Start the worker

docker compose up -d

4. Capture your session

In Claude, call manual_login — it extracts cookies from your logged-in Chrome tab and saves the session. You only need to do this once (or after your session expires).


Connect to Claude Code

Add to your MCP config (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "upwork": {
      "command": "node",
      "args": ["/absolute/path/to/upwork-mcp/dist/gateway.js"],
      "env": {
        "WORKER_PORT": "47821"
      }
    }
  }
}

Or via CLI:

claude mcp add upwork node /absolute/path/to/upwork-mcp/dist/gateway.js

Local Development (Hot Reload)

npm install
npm run worker   # starts tsc --watch + nodemon on dist/worker.js

The gateway (dist/gateway.js) runs separately and proxies to the local worker — changes to any tool file reload the worker instantly without restarting Claude.


Recommended Agent Workflow

1. manual_login      → Capture session (first time / session expired)
2. get_profile       → Check connects balance before bidding
3. search_jobs       → Find relevant jobs by keyword
4. get_job_details   → Get full description + screening questions
5. analyze_job       → Score the opportunity (skip grade D or F)
6. submit_proposal   → Auto-bid with personalized cover letter
7. get_messages      → Check for client responses
8. send_message      → Reply to clients
9. get_proposals     → Track active proposals

Example agent prompt

Search for freelance jobs matching my skills posted in the last 3 days.
For each job with grade A or B:
1. Get full job details
2. Analyze the opportunity
3. Write a personalized proposal highlighting my relevant experience
4. Submit at the recommended bid rate

Job Analysis Scoring

The analyze_job tool scores each job across 5 dimensions:

Dimension Max Points What It Measures
Niche Fit 30 Keyword match between job and your FREELANCER_NICHE
Client Quality 25 Rating, total spend, hire rate, payment verified
Budget Fit 20 Budget vs your target rate, estimated project value
Competition 10 Number of existing proposals (fewer = better)
Project Clarity 10 Description detail, skills listed, budget specified
Red Flags -5 each Vague scope, low-budget signals, no client history

Grades:

Grade Score Action
A+ 90–100 APPLY NOW — high priority
A 75–89 APPLY — strong opportunity
B 60–74 APPLY — worth pursuing
C 45–59 CONSIDER — marginal fit
D 30–44 SKIP
F <30 AVOID

Project Structure

upwork-mcp/
├── src/
│   ├── gateway.ts                # MCP stdio gateway (thin proxy, never restarts)
│   ├── worker.ts                 # HTTP tool server (hot-reloads via nodemon)
│   ├── config.ts                 # Environment configuration
│   ├── browser/
│   │   ├── browser-manager.ts    # Playwright CDP connection manager
│   │   └── upwork-auth.ts        # Session management
│   └── tools/
│       ├── search-jobs.ts
│       ├── get-job-details.ts
│       ├── analyze-job.ts        # Scoring engine
│       ├── submit-proposal.ts
│       ├── get-proposals.ts
│       ├── get-messages.ts
│       ├── send-message.ts
│       ├── get-profile.ts
│       ├── update-profile.ts
│       └── manual-login.ts       # CDP cookie extractor
├── cdp-proxy.cjs                 # Host-side proxy: Docker → Chrome (fixes Host header)
├── connect-chrome-docker.bat     # Launch Chrome + CDP proxy (for Docker use)
├── connect-chrome.bat            # Launch Chrome only (for local use)
├── Dockerfile
├── docker-compose.yml
└── .env.example

Notes

Session Management

Login is handled manually via manual_login. Open Chrome, log in to Upwork normally (including any 2FA), then call manual_login — it captures all cookies and saves them for Playwright to reuse.

Bot Detection

Upwork may detect unusual activity. Use reasonable request intervals and do not run the agent continuously. The server connects to your real Chrome profile via CDP, which behaves more like a real browser than headless automation.

Selector Stability

Upwork's UI changes periodically. Browser-based selectors in src/tools/ may need updating if the UI is redesigned.

Ethical Use


License

MIT

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
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
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
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
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
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
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
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