Overview

The Lancer MCP server lets an AI agent operate Lancer on your behalf through the Model Context Protocol. Once connected, your agent can:

• Configure your knowledge base, the profile the suitability AI reads when deciding which jobs fit you.
• Create and tune campaigns: filters, search queries, bidding and boosting config, go-live.
• Write and simulate cover letter templates with the same agent that writes your production proposals.
• Manage your inbox and pipeline: triage rooms, update CRM fields, create tasks, send messages, schedule meetings.
• Analyze performance and export leads or conversations for deeper analysis.
• Manage integrations: webhooks, Public API keys, Slack and Discord notifications, external mode.

The only thing that stays in the web app is connecting your Upwork account.

Using the MCP is the preferred way to set up and run Lancer.

Opinionated by design

A generic API integration leaves every judgment call to the agent. The Lancer MCP does not: alongside its 59 tools it ships Lancer's own operating knowledge, so any agent, on any model, behaves like an operator trained on how Lancer is meant to be run.

• 17 policy resources. Playbooks and rules the agent is required to read before acting: how to structure campaigns, how to write search queries, how each knowledge base field should be authored, what a good recommendation looks like. Browse them under Resources.
• Recommendation guardrails. Dedicated validation tools check every proposed campaign or knowledge base change against evidence requirements (real lead samples, cited examples, sufficient data) before the agent may present it to you. Ungrounded advice gets revised or blocked.
• Server-side enforcement. The highest-impact rules are not advisory. Writes that silently degrade lead quality, like removing budget floors or client-quality filters, are rejected by the server unless you explicitly opt in.
• Two guided prompts. Ready-made workflows for first-time setup and suitability tuning, so the most important journeys are scripted rather than improvised.

The practical effect: an agent cannot talk you into configurations that hurt your results, and the quality of its advice does not depend on it already knowing how Upwork works.

How it works

• Transport: Streamable HTTP, stateless. One endpoint, no session setup.
• Auth: a personal token embedded in the connect URL. See Connect your client.
• Scope: the token is tied to you, not to one organization. Tools accept an organizationId parameter, and your live organization memberships are checked on every call.

Where to start

First-time setup with an agent

If your organization is new, ask your connected agent to set Lancer up for you. The server ships a set_up_lancer prompt that walks the agent through profile grounding, knowledge base, campaigns, testing the AI's decisions, cover letters, and go-live.