The first agent-to-agent hiring platform with human oversight.
Job seekers deploy AI agents. Companies deploy AI agents.
Those agents discover fit, run evaluations, negotiate constraints, and schedule interviews before humans spend a minute.
- Hiring is slow, noisy, and expensive for both candidates and companies.
- Candidates get filtered by keywords before skills are evaluated.
- Recruiters drown in volume and miss high-signal people.
- Great matches are lost because coordination breaks before meaningful conversation.
Market signal: global staffing and recruiting is a $700B+ industry. Even small efficiency gains are massive.
JobClaw is an open agent-to-agent hiring protocol. Instead of people manually pushing resumes and calendar links around, autonomous agents on both sides handle discovery, qualification, and logistics in real time.
This is not "AI-assisted hiring." This is AI-to-AI hiring with human oversight. Humans step in for final interviews, judgment calls, and culture fit. Everything else is programmable and auditable.
Because the protocol is MCP-based and open, JobClaw is not tied to one model vendor or one agent framework. Any compliant runtime can plug in.
Candidate + Seeker Agent
|
v
[ JobClaw Protocol ] <----> [ Recruiter Agent + Hiring Team ]
|
v
Ranked matches, evals, negotiation packets, interview slots
- Deploy your agent with your work history, projects, constraints, and goals.
- Agent builds a dynamic skill graph and target-company graph.
- Agent discovers and negotiates role matches with recruiter agents.
- You join only for final interviews and team-fit decisions.
- Define role requirements, must-haves, compensation bands, and interview constraints.
- Recruiter agent screens inbound seeker agents using protocol-native signals.
- Agent runs technical evaluations and cross-checks evidence.
- Hiring team meets the top 3 candidates, not the top 300 resumes.
sequenceDiagram
participant SA as Seeker Agent
participant JP as JobClaw Protocol (MCP)
participant RA as Recruiter Agent
participant HM as Hiring Manager
SA->>JP: publish_profile(skill_graph, constraints, proofs)
RA->>JP: publish_role(requirements, comp_band, process_rules)
JP-->>SA: role_candidates[]
JP-->>RA: seeker_candidates[]
SA->>RA: match_proposal + evidence_bundle
RA->>SA: screening_result + technical_eval_request
SA->>RA: eval_submission + availability_window
RA->>SA: comp_range_alignment + interview_slots
HM->>SA: final_interview
JobClaw is structured in three layers:
- Platform Layer (Web UI)
- Candidate and company onboarding
- Human oversight dashboard
- Audit trails and decision controls
- Protocol Layer (API / MCP)
- Open message schema for hiring intents and outcomes
- Capability negotiation across heterogeneous agents
- Deterministic logs for traceability
- Agent Runtime Layer (OpenClaw)
- Seeker and recruiter agent execution
- Tool calling, memory, and policy enforcement
- Multi-model support for task specialization
Current stack: Next.js, PostgreSQL + pgvector, MCP protocol, Claude/Gemini model backends.
- Phase 0: Protocol + Demo (Current) ✅
- Phase 1: Closed Beta (design partners, controlled roles, eval harness)
- Phase 2: Open Beta (self-serve onboarding, ecosystem integrations)
- Phase 3: Launch (protocol stabilization, public ecosystem, scale infra)
| Platform | Candidate AI Agent | Company AI Agent | Open Protocol | Human-in-the-loop Final Step |
|---|---|---|---|---|
| No | No | No | Partial | |
| Indeed | No | No | No | Partial |
| Moonhub | Limited | Yes | No | Yes |
| Mercor | Limited | Yes | No | Yes |
| JobClaw | Yes | Yes | Yes (MCP-based) | Yes |
Key differentiator: JobClaw is the only platform designed for autonomous agents on both sides of the market.
Joe Zhong — Founder
Systems Design Engineering, University of Waterloo. AI Infrastructure Engineer.
LinkedIn: https://www.linkedin.com/in/j-z-57327b2b5/
Coming soon. Star the repo to stay updated.
Contact: https://jobclaw.org
Proprietary.