Dice Job Search Workflow

Dice Job Search Workflow

Use the bundled Dice remote MCP config as a read-only job-search data source for technology roles.

This skill helps the agent turn a job-search goal into bounded Dice MCP searches, compare returned jobs against the user's stated preferences, and produce useful next actions without taking over the application or profile lifecycle.

Inputs

• Required: the role target, search question, or job-search goal
• Useful: location, remote/hybrid/on-site preference, employment type, recency, visa sponsorship needs, salary expectations, dealbreakers, target stack, resume or profile excerpt, and maximum number of roles to review
• Optional: existing Dice profile state, saved searches, application tracker, preferred resume variant, and whether recurring search automation is desired

Source

Professional Skills bundles a plugin-provided MCP config for Dice at plugins/professional-skills/.mcp.json. Dice documents a remote MCP endpoint at https://mcp.dice.com/mcp with a read-only search_jobs tool. No Dice login is required for basic public job search. Treat this as an external live data source whose listings, filters, and client setup details may change.

Official sources:

• Dice MCP technical documentation: https://www.dice.com/about/mcp
• Dice MCP setup guide: https://www.dice.com/career-advice/how-to-connect-the-dice-mcp-server-to-your-ai-assistant
• Dice launch article: https://www.dice.com/career-advice/dice-launches-mcp-server-for-ai-powered-job-search

Workflow

1. Clarify the search goal in practical terms: target role, geography, work mode, seniority, technologies, employment type, recency, and hard filters.
2. Verify whether the plugin-provided Dice MCP server is enabled and available in the active session.
   • If available, use it directly for read-only search.
   • If the plugin is installed but Dice is disabled, tell the user to enable the plugin-provided dice MCP server rather than hand-editing user config first.
   • If unavailable, give setup or handoff guidance using the official endpoint and do not invent local server code.
3. Prefer broad first searches, then narrow with filters after seeing the shape of results. Avoid overfitting the first query to every preference.
4. Use Dice's documented parameters when constructing searches:
   • keyword: required job title or search keywords
   • location: city, state, region, or remote-oriented search term
   • radius and radius_unit: location expansion
   • workplace_types: Remote, On-Site, or Hybrid
   • employment_types: FULLTIME, CONTRACTS, PARTTIME, or THIRD_PARTY
   • employer_types: Direct Hire, Recruiter, or Other
   • posted_date: ONE, THREE, or SEVEN
   • willing_to_sponsor: work authorization sponsorship filter
   • easy_apply: simplified application filter
   • jobs_per_page, page_number, and fields: pagination and response size
5. Review results for role fit before recommending action. Separate:
   • strong matches worth applying to
   • possible matches worth a resume variant or follow-up search
   • poor matches blocked by hard constraints
   • ambiguous matches needing the full job page, current company info, or user decision
6. When comparing against a resume, keep the output evidence-heavy:
   • match required skills to resume proof
   • name missing or weak proof
   • suggest concrete resume edits only when grounded in real experience
   • avoid keyword stuffing, invented experience, and broad ATS filler
7. For recurring searches or digests, design the search prompt and cadence first. Require explicit user approval before creating automations, notifications, application trackers, or any write surface outside the current answer.
8. If the user wants to apply, keep the boundary clear: prepare materials, summarize risks, and ask before submitting, changing profile data, or sending messages.

Search Patterns

Use these as starting shapes, not canned final answers:

• Broad discovery: keyword, optional location, jobs_per_page: 25
• Recent remote search: keyword, workplace_types: ["Remote"], posted_date: "THREE"
• Local hybrid search: keyword, location, radius, radius_unit: "mi", workplace_types: ["Hybrid"]
• Sponsorship search: keyword, willing_to_sponsor: true, plus broad location or remote filters
• Recruiter-light search: search normally, then classify employer type and recruiter involvement from the returned fields when available
• Skills trend check: run a small set of comparable searches, sample results, and summarize repeated technologies without claiming full-market statistics

Output Contract

For job triage, return:

• Answer: the direct recommendation or shortlist
• Meaning: why the roles or search shape fit the user's goals
• Risk: hard blockers, stale/live-data uncertainty, and decisions needed
• Evidence: search parameters, result counts, listing names, dates, and links
• Next Search: one focused follow-up query when useful

For setup guidance, return:

• Answer: whether Dice MCP is currently usable in this environment
• Meaning: what the endpoint/tool enables
• Risk: authentication, subscription, client, rate-limit, and privacy caveats
• Details: client-specific setup steps and verification prompt

Guardrails

• Do not apply to jobs, send messages, alter a Dice profile, upload resumes, or create saved searches unless the user explicitly asks for that exact action.
• Do not claim a role is current, remote, sponsor-friendly, or easy-apply unless that field is present in live Dice results or the visible listing evidence.
• Do not infer salary, sponsorship, employer type, or remote status from vague wording when the result does not expose it.
• Do not promise that Dice MCP stores no data beyond what the official docs say. State only that the documented server exposes public job listings and basic search does not require a Dice login.
• Do not scrape Dice web pages when the MCP tool is available and sufficient.
• Do not build a local MCP server, proxy, cache, or automation runner inside this skill. This skill is guidance around Dice's external MCP surface.
• Respect rate limits. If HTTP 429 or throttling appears, stop, back off, and report the limit instead of retrying aggressively.
• Keep personal job-search preferences, resumes, and application history out of logs and durable repo files unless the user explicitly asks to save them.

References

• agents/openai.yaml
• references/dice-mcp-source-notes.md