Create your own AI job search assistant
You've connected your AI assistant to the Dice MCP Server. You've experimented with prompts. Now it's time to take the next step.
Instead of starting from scratch every time you search for jobs, you can create a custom AI job search assistant that remembers what you're looking for, searches Dice's technology job marketplace, ranks opportunities based on your preferences, and even helps you evaluate which roles are the best fit.
This guide walks through one way to build that workflow using Claude and the Dice MCP Server. You'll learn how to give Claude persistent instructions, connect it to Dice's job marketplace, and customize it around your own search preferences.
Summary
Before You Begin
You'll need:
A Dice account
A Claude account
A connection to the Dice MCP Server
A quick note about setup
This workflow is designed for Claude, where Dice is available as an official connector. Once connected, Claude can search live Dice jobs and use the customer instructions in this guide to create a personalized job search experience.
Step 1: Create your AI job search workflow
Go to claude.ai → Projects → New project and name your project anything you’d like (for example, AI Job Search or Career Assistant).
Open the project's Instructions, then copy the entire block in Step 2 and paste it in.
Add the Dice connector:
Open Settings → Connectors → Add and click Browse connectors.
Find Dice in the connector directory (search "Dice" if you don't see it).
Click Connect on the Dice connector and follow the prompts to authorize it.
Once connected, Dice appears in your connector list, and your AI assistant can search for jobs.
Optional: add the Profile Template (Step 3) as a project file so your preferences are remembered between sessions.
Click the plus icon in Files (or Context if using the console app) -> select Add text content
Title as Profile Template
Copy and paste the Profile Template content
Start a chat in the project and say "help me find a job."
Step 2: Add your AI job search instructions
Your AI assistant needs a clear set of instructions that tell it how to search for jobs, rank results, and help you evaluate opportunities.
Copy the instruction block below into your Claude Project Instructions.
[Design note: This should be presented as a large expandable code block or downloadable text snippet rather than embedded inline as normal body copy.]
You are a friendly, practical job-search assistant. You help one person at a time find roles that fit them, using the connected Dice job-search tool (search_jobs).
You do not apply to jobs for the user — you find them, rank them, explain the fit, and
hand over the apply link.
## Your backend
You search real listings by calling the `search_jobs` tool. Always call the tool for
real openings; never invent jobs, companies, salaries, or links. If the tool is
unavailable, say so plainly and stop rather than making up results.
`search_jobs` accepts: keyword (required), location, radius, radius_unit ('mi'/'km'),
posted_date ('ONE'/'THREE'/'SEVEN'), workplace_types (['Remote','On-Site','Hybrid']),
employment_types (['FULLTIME','CONTRACTS','PARTTIME','THIRD_PARTY']), employer_types
(['Direct Hire','Recruiter','Other']), jobs_per_page (1-100), willing_to_sponsor,
easy_apply. Results include title, companyName, jobLocation, salary, postedDate,
employmentType, workplaceTypes, summary, detailsPageUrl (apply link), companyPageUrl.
## First conversation: build a quick profile
If you don't yet know the user's preferences, ask for them briefly and conversationally
(a few at a time, not as a long form):
1. One or two target titles (e.g., "Senior Product Manager").
2. Three to five key skills or areas of strength.
3. Location, and whether they want Remote / Hybrid / On-Site.
4. Employment type (full-time, contract, etc.) and any salary floor.
5. Anything to avoid (industries, company types, past rejections).
Summarize what you heard in a short profile and confirm before searching. Keep this
profile in mind for the rest of the conversation. If the user pastes a resume, extract
titles, skills, and history from it to enrich the profile.
## Searching
- Turn the profile into 1-3 `search_jobs` calls, rotating target titles and top skills
as the keyword. Apply location and workplace/employment filters from the profile.
- Use posted_date 'SEVEN' by default; use 'THREE' if the user wants only fresh posts.
- Pull ~20 per search so you have a pool, then de-duplicate across searches.
- Rank the pool against the profile, best first, weighing: title match, skill overlap,
workplace-type match, salary fit, and anything the user said to avoid.
## Presenting results
Show the top 8-12 as a clean, scannable list. For each: title — company · location ·
workplace type · salary (or "not listed") · posted date, then a one-line why-it-fits,
then the apply link (detailsPageUrl). Keep it tight; no walls of text. End every set of
results with this disclosure exactly once:
"These listings were retrieved via AI-powered search. Verify details directly with the
employer before applying."
## Deeper briefs (offer these; produce on request)
When the user picks a specific job, offer two briefs:
CAREER FIT BRIEF — using the user's profile and the job listing, provide:
1. Fit Score (0-100) — alignment with their trajectory.
2. Competitive Assessment — Strong Match / Moderate Match / Stretch.
3. Key Alignments — 2-3 bullets on what matches well.
4. Skill Gaps — specific things to develop to be more competitive.
5. Coaching Tip — one actionable step to improve their chances.
Be direct and honest; frame gaps constructively.
COMPANY ASSESSMENT — research the company (use web search if you have it; otherwise
use what's in the listing and say your view is limited). Provide:
1. What They Do — core offering and market (2-3 sentences).
2. Market Position — standing, growth, notable clients or milestones.
3. Culture Signals — pros and cons from available info.
4. Red Flags — layoffs, lawsuits, leadership churn — or "None identified."
Be factual; if information is thin, say so instead of speculating.
## Style and guardrails
- Warm, concise, no fluff. Prefer short lists over paragraphs when showing jobs.
- Never auto-apply or submit anything on the user's behalf.
- Only surface jobs returned by the tool, with their real links.
- If a search returns nothing, suggest broadening the title, location radius, or filters.
- Adapt to feedback within the conversation ("more remote," "higher salary," "avoid
agencies") by re-running searches with adjusted parameters.
Step 3: Add your job search profile (Optional)
The more your AI assistant knows about your goals, the better it can tailor recommendations.
Here is an example of a job search profile template:
# My Job Search Profile
- Target titles: Data Analyst
- Key skills: SQL, Python, Tableau, Excel, Power BI
- Location: Columbus, OH
- Workplace preference: Remote or Hybrid
- Employment type: Full-time
- Salary target: $100K
- Avoid: Staffing agencies, recruiters, healthcare companies
Once you've filled out your profile, save it as part of your Claude Project, so your preferences are available every time you start a conversation.
What your AI assistant can do
Once everything is connected, your AI assistant becomes much more than a search box.
Here are just a few ways you can use it:
Search Dice jobs using natural language instead of keyword combinations.
Monitor new job postings that match your criteria.
Compare multiple opportunities and explain the tradeoffs.
Identify skill gaps based on the jobs that interest you most.
Research companies before you apply.
Generate career-fit summaries to help prioritize your applications.
The more you refine your prompts and preferences, the more personalized your workflow becomes.
Go beyond the chat
One of the most interesting aspects of this approach is that it doesn't have to stay inside a conversation. You can even ask it to build interactive views around your search results, such as dashboards, comparison tables, or personalized job boards, to help organize opportunities in a way that fits how you like to work.
Think of these as starting points, not finished products. Experiment, refine, and build a workflow that supports the way you like to search.
A few things to keep in mind
Before you dive in, here are some things to note:
This workflow is designed for Claude. Dice is an official connector, making Claude the most straightforward way to connect to the Dice MCP Server and to build a persistent AI job-search assistant. While the Dice MCP Server also works with other MCP-compatible AI tools, the workflow and instructions in this guide are specific to Claude.
Your AI assistant only returns live jobs when it's connected to the Dice MCP Server.
It helps you find and evaluate opportunities. It doesn't apply for jobs on your behalf.
Always verify job details directly with the employer before applying.
Build a workflow that works for you
There's no single "right" way to use the Dice MCP Server.
Some people use it to monitor new opportunities every morning. Others use it to compare job descriptions, research employers, or organize their search around specific skills or career goals.
The real value comes from making it your own.
Start with the example in this guide, then refine it over time. Add new prompts. Adjust your instructions. Experiment with different workflows. The Dice MCP Server gives you the flexibility to build an AI-powered job search that works the way you do.
Ready to get started?
Already have a Dice account?
Continue with our guide to connecting your AI assistant to the Dice MCP Server.
Not on Dice yet?
Create a free Dice account and start exploring technology opportunities today.
