AGENT

Agent Integration Guide

WPORT is more than an API — it ships the mechanisms that let AI agents discover, read and operate the platform on their own. Here are the four core pieces.

① .well-known discovery chain

From the API catalog an agent discovers the OpenAPI schema and Agent Skills — no hard-coded endpoints:

bash
curl https://wport.me/.well-known/api-catalog
#  → service-desc: /.well-known/openapi.json

② Agent Skills

wport-agents ships pluggable skills that move an agent from answering to actually getting things done:

interviewer-aiMock technical interviews with instant feedback
gen-career-mentorCareer-path advice based on your skills
exec-wport-cliSkill drives the CLI to fetch data
gen-resumeGenerate a resume from a conversation
gen-resume-optimizerOptimize a resume for a job

Each skill is a SKILL.md that an AI agent (e.g. Claude) loads to use. See the full list and usage in the wport-agents repository: github.com/hotfire-digital/wport-agents

③ WebMCP browser tools

The WPORT site registers tools in the browser via the W3C WebMCP standard (navigator.modelContext). These tools are navigational — they take the browser to the right page and return the current page context (path / url / title); they are not data-returning query APIs:

javascript
// app/plugins/webmcp.client.ts
navigator.modelContext.provideContext({ tools })
//  → open-jobs-list({ keyword, page })
//  → open-job-detail({ id })
//  → open-company-detail({ id })
//  → open-auth({ tab, redirect })
//  → get-current-page-context()

On the server side, an MCP discovery card at /.well-known/mcp/server-card.json advertises the streamable-http transport and capabilities to external MCP clients; the endpoint itself is POST /mcp, exposing 8 read-only tools (searchJobs, getPopularJobs, getJob, searchCompanies, getPopularCompanies, getCompany, getCompanyJobs, listCourses) whose names match the operationIds in openapi.json. No authentication required.

This endpoint is not the same thing as the two OAuth connectors on mcp.wport.me: POST /mcp here needs no authentication and only serves public data. Reading your own resumes, or your company's jobs and talent pool, goes through the authorized job-seeker and employer lines instead.

④ Markdown content negotiation

Send Accept: text/markdown and pages return LLM-friendly plain-text Markdown — no HTML parsing needed:

bash
curl -H "Accept: text/markdown" https://wport.me/jobs

Call the agent endpoint

For programmatic integration, /site-api/agent provides an agent-oriented entry point; the schema is always defined by .well-known/openapi.json.

See also