Prompts
Ready-made prompts that put your agent into a guided Lancer workflow.
Prompts are ready-made instructions shipped by the Lancer MCP server. Invoking one hands your agent a complete, tested playbook, so you do not have to explain Lancer to your agent yourself.
There are three ways to run one:
- Native MCP prompts. Clients with prompt support list them once the server is connected. In Claude Code they appear as slash commands, in Claude Desktop under the prompts picker in the chat input.
- Copy from this page. Each prompt below comes with a paste-ready version; use the copy button on the block.
- Copy from the web app. The Prompts card on the MCP page in the Lancer web app has the full text of every prompt with a click-to-copy.
All routes lead to the same behavior, so use whichever is more convenient.
Available prompts
set_up_lancer
Guided first-time setup: knowledge base, campaigns, cover letter, and go-live, grounded in your Upwork profile.
In Claude Code, type:
/mcp__lancer__set_up_lancer(The lancer part is whatever name you gave the server when connecting; the same pattern works for any prompt.)
In any other client, copy the full prompt and paste it into a chat with Lancer connected:
Use it when your organization is new, or when you want to redo your setup properly. The agent walks you through six phases and confirms each step with you before writing anything:
Profile grounding. The agent reads your connected Upwork account with bidder_account_get. If you have not connected one yet, it scrapes your public Upwork profile by URL instead, and only asks you to describe your positioning manually as a last resort.
Knowledge base. It drafts a knowledge base from your profile with knowledge_base_draft, then reviews it with you field by field: core services, tools and technologies, projects you avoid, languages, case studies.
Campaigns. It proposes one or more campaigns with filters and search queries that match your positioning, checked against Lancer's campaign policy before anything is saved.
Test the AI's decisions. It runs knowledge_base_simulate on real jobs so you can see which jobs Lancer would accept or reject for you, and tunes the knowledge base until the decisions look right.
Cover letter. It creates a template and offers cover_letter_simulate, which renders what the production proposal agent would actually write for sampled jobs.
Go-live. Final checks (connected bidder account, startability requirements), then the campaign is activated with your confirmation.
Expect the full flow to take 15 to 30 minutes of conversation. You can stop at any phase and resume later; every phase leaves Lancer in a consistent state.
tune_suitability
Benchmark the suitability AI on a few hundred real jobs, grade the contested decisions together, and iterate the knowledge base until it matches your judgment.
In Claude Code, type:
/mcp__lancer__tune_suitabilityIn any other client, copy the full prompt and paste it into a chat with Lancer connected:
Use it once your knowledge base exists and you want to trust the AI's accept/reject decisions before (or after) going live. It is the deep, standalone version of the setup flow's testing phase:
Full-context grounding. The agent loads everything the decision should be judged against: your Upwork title, description, skills, past jobs, portfolios, and the complete knowledge base including case studies.
Baseline sweep. It runs knowledge_base_simulate over roughly 200 recent real jobs (you confirm the scale first; each job is one AI call), going deeper when the agent judges more signal is needed, and records every accept/reject decision.
Triage. The agent grades every decision itself against your profile and knowledge base, and sorts them into agree, disagree, and unsure.
You grade the contested ones. Only the disagreements and borderline cases come to you, each with the real job post quoted, in digestible batches. Your verdicts are ground truth.
Knowledge base iteration. Confirmed mistakes become targeted knowledge base patches, following the same authoring rules and guardrails as any other edit, verified by re-simulation before anything is saved.
The loop targets roughly 95% agreement and is capped at 3 iterations; it will tell you which residual disagreements are genuine judgment calls rather than fixable mistakes. Expect it to take a while: the baseline sweep alone is several minutes of simulation.
More prompts
This catalog will grow. Prompts for campaign tuning, weekly performance reviews, and pipeline cleanup are planned. If there is a workflow you keep explaining to your agent by hand, that is a good candidate; tell us through the app.