MCP · Open standards · Agent infrastructure

OJCP — The Open Standard for Agent-Consumable Job Feeds

OJCP defines how AI agents discover, reason over, and act on job opportunities. Built on MCP. Interoperable with schema.org. Governed by an independent steering committee.

By Ashrey

A new open protocol launched this week that could fundamentally change how AI agents interact with the hiring market. OJCP (Open Job Context Protocol) is a v0.1 draft specification that defines a standard way for AI agents to discover, reason over, and act on job opportunities — without human operators manually copying job descriptions between systems.

What OJCP Actually Is

OJCP is not a job board. It's a protocol layer that sits between job data sources (ATS platforms, job boards, company career pages) and AI agents that want to consume that data. The spec defines:

  • Manifest files — a .well-known/ojcp.json that advertises an employer's job feed location and supported capabilities
  • Job feed schema — a structured JSON format extending schema.org JobPosting with agent-specific fields (skills as structured entities, compensation ranges, remote policies, visa sponsorship signals)
  • MCP tools — four standard tools (search_jobs, get_job_detail, get_employer_context, begin_application) that any compliant server implements
  • Apply paths — standardized ways for agents to initiate applications, from simple "apply with profile" to multi-step verification flows

The protocol is built on MCP (Model Context Protocol), meaning any MCP-compatible agent (Claude Code, Codex, custom agents) can call OJCP tools natively without custom integration code. It also aligns with schema.org vocabulary so existing structured data on career pages can map cleanly into OJCP feeds.

Why This Matters for Developers Building Agents

If you're building an AI agent that helps users with career moves — whether that's a job-search assistant, an application autofiller, or a recruiting agent — OJCP eliminates the brittle scraping layer. Today, every agent builder writes their own Greenhouse parser, their own Lever scraper, their own LinkedIn HTML parser. OJCP replaces all of that with a single MCP tool interface.

The spec also solves the "context problem" for agents. A raw job posting lacks the employer context an agent needs to reason well: company size, tech stack, team structure, interview process, culture signals. OJCP's get_employer_context tool returns a structured EmployerContext object that agents can use to tailor applications, prepare candidates, or filter opportunities.

Governance and Adoption

OJCP is governed by an independent steering committee — not a single vendor. The spec is CC BY 4.0, reference implementations are Apache 2.0. Founding members include Workday, CrossCountry, Recruitics, LoopCV, HiringCafe, AIApply, ScaleJobs, and Tink — a mix of ATS platforms, job boards, and AI hiring tools.

This governance model matters. If OJCP were owned by one ATS vendor, competing platforms would have no incentive to adopt. The independent committee and open license create a neutral layer where everyone benefits from interoperability.

Current State: v0.1 Draft

The spec is explicitly labeled v0.1 Draft. The full specification covers tools, schemas, apply paths, and security requirements. There's a live playground where you can execute MCP tools against a mock registry (Acme Corp, Globex, etc.) to see the request/response shapes.

Reference implementations are early. The GitHub org ojcp-org hosts schemas, examples, and a TypeScript SDK. Community discussions happen in GitHub Discussions — the steering committee is actively seeking implementer feedback before v1.0.

What to Watch

Three signals will determine whether OJCP becomes the de facto standard or stays a niche experiment:

  1. ATS adoption — Will Greenhouse, Lever, Ashby, Workday publish .well-known/ojcp.json manifests? Without major ATS platforms, the feed coverage stays thin.
  2. Agent framework support — Will MCP client libraries (Python, TypeScript, Go) add first-class OJCP helpers? The protocol is only as useful as the ergonomics of calling it.
  3. Candidate privacy model — The spec includes CandidateContext for agents to share candidate data with employers. Getting the privacy/consent UX right is non-trivial and will face scrutiny.

For now, OJCP is the most credible attempt yet at a universal "job feed for agents" standard. If you're building in the AI + hiring space, it's worth reading the spec and experimenting with the playground. The protocol design is clean, the governance is neutral, and the MCP foundation means integration cost is low for any agent already speaking MCP.

Everything the lab builds in public stays in public. The source is on GitHub, and the current build is a live AI quoting engine for home service contractors.

← All posts