upwork-mcp
An MCP server for automating Upwork workflows including job search, proposal submission, client communication, and contract management. It provides tools for client vetting, template-based proposals, and session safety with audit logging.
README
upwork-mcp
An MCP server for Upwork automation. Search jobs, draft and submit proposals, manage messages and contracts — all from Claude Desktop or any MCP-compatible client.
Features
- Job Search — keyword search with filters (budget, hourly rate, experience level, client rating, recency), automatic scoring and red-flag detection
- Proposal Workflow — draft from templates, preview, then submit with explicit confirmation; daily limit enforced
- Client Vetting — 7-condition red-flag checker (payment verification, rating, hire history, account age, repost count, proposal volume, hourly rate floor)
- Messages — list unread conversations, read threads, draft and send replies
- Contracts — list active/past contracts, view details and work diary entries
- Session Safety — persistent Chromium profile, Cloudflare detection, 7-day idle expiry, domain restriction to
upwork.com - Audit Log — every tool call appended to
~/.upwork-mcp/audit.log(metadata only, no content)
Requirements
- Python 3.11+
- uv (recommended) or pip
- A valid Upwork account
Installation
Using uv (recommended)
git clone https://github.com/kubegrind/upwork-mcp
cd upwork-mcp
uv sync
uv run playwright install chromium
Using pip
git clone https://github.com/kubegrind/upwork-mcp
cd upwork-mcp
pip install -e .
playwright install chromium
Configuration
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"upwork": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/upwork-mcp",
"run",
"upwork-mcp"
]
}
}
}
Claude Code
claude mcp add upwork -- uv --directory /absolute/path/to/upwork-mcp run upwork-mcp
Environment Variables
| Variable | Default | Description |
|---|---|---|
UPWORK_MCP_HOME |
~/.upwork-mcp |
Data directory for profile, logs, templates |
UPWORK_MCP_MIN_HOURLY_RATE |
25 |
Minimum acceptable hourly rate for red-flag detection |
Authentication
The server uses a persistent Chromium browser profile. On first run, authenticate manually:
uv run upwork-mcp --no-headless
A browser window will open. Log in to Upwork normally. The session is saved to ~/.upwork-mcp/profile/ and reused on subsequent runs. Sessions expire after 7 days of inactivity.
Tools
Jobs
| Tool | Description |
|---|---|
upwork_search_jobs |
Search by keyword with filters; returns scored, ranked results |
upwork_get_job_details |
Full job details including client history and interview rate |
upwork_score_job |
Score a specific job 1–10 |
upwork_flag_client |
Check a client against 7 red-flag conditions |
upwork_run_saved_search |
Run a saved search from ~/.upwork-mcp/searches/<name>.json |
Proposals
| Tool | Description |
|---|---|
upwork_get_proposals |
List submitted proposals with status |
upwork_get_proposal_details |
Details for a specific proposal |
upwork_draft_proposal |
Draft a proposal from a template |
upwork_preview_proposal |
Preview a draft before submission |
upwork_submit_proposal |
WRITE — submit a proposal (requires confirmed: true) |
upwork_withdraw_proposal |
WRITE — withdraw a submitted proposal (requires confirmed: true) |
Messages
| Tool | Description |
|---|---|
upwork_get_messages |
List unread conversations |
upwork_get_conversation |
Read the last 10 messages in a thread |
upwork_draft_message |
Draft a reply for review |
upwork_send_message |
WRITE — send a message (requires confirmed: true) |
Contracts
| Tool | Description |
|---|---|
upwork_get_contracts |
List active and past contracts |
upwork_get_contract_details |
Contract details and milestones |
upwork_get_work_diary |
Hours and memo entries for a contract |
Profile
| Tool | Description |
|---|---|
upwork_get_my_profile |
Name and title from your profile |
upwork_get_connects_balance |
Current connects balance with low-balance warning |
upwork_get_profile_stats |
JSS, Top Rated status, and total earnings |
Session & Audit
| Tool | Description |
|---|---|
upwork_check_session |
Verify the session is authenticated |
upwork_close_session |
WRITE — close the browser session |
upwork_rotate_session |
WRITE — restart the browser context |
upwork_view_audit_log |
View the last N audit log entries |
Proposal Templates
Templates live in ~/.upwork-mcp/templates/. The default template is loaded from default.txt if it exists. Pass a filename via template_name to use a custom template.
Supported variables:
{job_title} {client_name} {key_skill}
{estimated_timeline} {rate} {years_experience}
{industries} {custom_paragraph} {start_date}
{your_name}
Example template (~/.upwork-mcp/templates/default.txt):
Hi {client_name},
I noticed you're looking for {key_skill} help on {job_title}.
I have {years_experience} years of relevant experience and can deliver {estimated_timeline}.
My rate is {rate}.
{your_name}
Saved Searches
Create JSON files in ~/.upwork-mcp/searches/ to save search configurations:
{
"query": "devops kubernetes",
"experience_level": "expert",
"budget_min": 500,
"client_rating_min": 4.5
}
Run with upwork_run_saved_search(search_name="devops").
Job Scoring
Jobs are scored 1–10 using a weighted model:
| Factor | Weight |
|---|---|
| Budget match | 30% |
| Skill match | 25% |
| Client quality | 20% |
| Proposal competition | 15% |
| Recency | 10% |
Jobs scoring below 4.0 are filtered from results. Jobs scoring 7.0 or above are marked [HIGH PRIORITY].
Safety
- WRITE tools (
submit_proposal,withdraw_proposal,send_message,close_session,rotate_session) requireconfirmed: true— the model cannot act without explicit user confirmation - Daily proposal limit — hard block at 10 submissions per day
- Domain restriction — browser navigation is restricted to
upwork.com - No credential exposure — cookies and tokens are never returned, logged, or printed
- Path traversal protection — template and saved search filenames are validated with
Path.is_relative_to() - Audit log — all tool calls recorded with timestamp, action type, target, and status
Development
uv sync --extra dev
uv run pytest
uv run ruff check src
uv run mypy src
License
MIT
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。