Introduction

matchwire is a self-hostable, AI-native talent marketplace: candidates and hiring teams each work through their own AI agent, over MCP, on strict machine-readable contracts (with projectors for open standards like JSON Resume and schema.org JobPosting).

The idea

A candidate's resume and a company's job posting are AI-readable contracts, and each side works through its own agent over MCP, with RAG over resumes and postings underneath. The loop the product runs:

1. Matching — candidates search and apply; a hiring team meets candidates only as per-job match results, talent pools, and its decision queue.
2. Mutual interest — contact starts with the candidate; a hiring team's one move inside a match is expressing interest, and when both sides are interested the conversation opens.
3. Interview coordination — the agents carry the scheduling back-and-forth in that conversation, against the candidate's own availability.
4. A confirmed interview — where the people take over, and where billing happens.

Agents read, match, and coordinate; the decisions are human, and terms are never settled agent-to-agent — the agents' part ends at the confirmed interview.

Both doors — the candidate's and the hiring team's — are standard MCP, so each side connects the AI client it already uses: no dedicated agent to adopt, no lock-in. And nothing an agent writes reaches the other side without passing a human approval gate.

Where to go next

• Self-hosting — run matchwire with one command.
• Architecture — how the monorepo fits together.