upwork-mcp-server
Open-source MCP server for Upwork's GraphQL API — search jobs, manage contracts, send messages, and more through AI agents like Claude
Install / Use
claude mcp add AbbottDevelopments -- npx -y github:AbbottDevelopments/upwork-mcp-serverIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
LegalSupported Platforms
Skill content
View source on GitHubUpwork MCP Server
An open-source Model Context Protocol server that connects AI agents to the Upwork API — enabling agentic job discovery, client intelligence, and freelancer workflow automation.
Why This Exists
Freelancers spend hours browsing job boards, evaluating fit, and writing proposals. AI agents are great at this kind of work — filtering, summarizing, drafting — but they need a standardized way to access the data.
This MCP server gives any MCP-compatible AI client (Claude Desktop, custom agents, etc.) direct access to Upwork's API through a clean, tool-based interface.
Features
- 15 agentic tools covering the full freelancer workflow: discover → score → propose → manage → communicate
- OAuth 2.0 authentication — Secure localhost callback flow with automatic token refresh
- Composite scoring —
score_opportunityrates jobs 0-100 across client quality, budget fit, competition, and skill match - GraphQL + REST — Uses Upwork's GraphQL API with REST fallbacks where needed
- Rate limiting — Token bucket (10/sec, 300/min, 40K/day) with exponential backoff
- Structured errors — Typed error categories with retryability flags and actionable messages
- Poll notifications — Stateful diffing for new messages, proposal updates, and job matches
Quick Start
Prerequisites
- Node.js 18+
- Upwork API credentials (developer portal)
Setup
# Clone and install
git clone https://github.com/AbbottDevelopments/upwork-mcp-server.git
cd upwork-mcp-server
npm install
# Configure credentials
cp .env.example .env
# Edit .env with your UPWORK_CLIENT_ID and UPWORK_CLIENT_SECRET
# Build
npm run build
# Authenticate (one-time OAuth flow)
npm run auth
# Run the server
npm start
Connect to Claude Desktop
Add to your Claude Desktop config (claude_desktop_config.json):
{
"mcpServers": {
"upwork": {
"command": "node",
"args": ["/path/to/upwork-mcp-server/dist/index.js"],
"env": {
"UPWORK_CLIENT_ID": "your-client-id",
"UPWORK_CLIENT_SECRET": "your-client-secret"
}
}
}
}
Connect to Claude Code
claude mcp add upwork -- node /path/to/upwork-mcp-server/dist/index.js
Tools
| Tool | Description |
|------|-------------|
| search_jobs | Search Upwork jobs with 15+ server-side filters |
| get_job_details | Full job posting with embedded client intelligence |
| score_opportunity | Composite 0-100 quality scoring with red/yellow/green flags |
| analyze_client | Deep client risk assessment from a job posting |
| get_my_profile | Your freelancer profile, skills, rate, and JSS |
| get_connects_balance | Available connects before applying |
| search_freelancers | Competitor research by skills, rate, JSS, country |
| track_proposals | List proposals with status and outcomes |
| list_contracts | Active/ended contracts with filtering |
| get_contract_details | Full contract with milestones and earnings |
| manage_milestones | Create, edit, or request approval on milestones |
| get_earnings_report | Earnings and time reports by period or custom range |
| list_messages | Message rooms with unread counts and previews |
| send_message | Send a message in an existing room |
| poll_notifications | Check for new messages, proposal updates, and job matches |
Architecture
See docs/architecture.md for diagrams showing how the system connects — also available as an Excalidraw file for interactive editing.
For implementation details, see the technical specification.
Project Structure
src/
index.ts # MCP server entry point
auth/
oauth.ts # OAuth 2.0 authorization flow
token-store.ts # Persistent token storage (~/.upwork-mcp/)
common/
api-client.ts # Upwork API client with auth + rate limiting
errors.ts # Structured error types
rate-limiter.ts # Request throttling
poll-store.ts # Polling state management
types.ts # TypeScript type definitions
graphql/
queries.ts # Read operations (jobs, profiles, contracts)
mutations.ts # Write operations (proposals, messages)
operations/ # High-level MCP tool implementations
Development
# Watch mode (rebuilds on change)
npm run dev
# Run tests
npm test
# Run tests in watch mode
npm run test:watch
Status
All 15 tools implemented — OAuth flow, API client, rate limiting, and all four phases (scaffold, core tools, work management, communication) are built and reviewed. Ready for live API testing with Upwork credentials.
Contributing
Issues and PRs welcome. If you're a freelancer who wants specific Upwork workflows exposed as MCP tools, open an issue describing your use case.
License
MIT — see LICENSE for details.
Author
Built by Abbott Developments — AI automation and process consulting for operations-heavy businesses.
Related Skills
Agent-Reach
76.5kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
69.7k🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
headroom
68.0kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
CowAgent
46.7kOpen-source super AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-model, multi-channel. Lightweight, extensible, one-line install. (formerly chatgpt-on-wechat)
