AI Agent Platform — Overview

JobMatch exposes a curated tool API so companies can connect their own LLM agents — built on Claude, GPT, Gemini, Llama, or any MCP-capable client — to operate on the platform. Your agent can manage jobs, search candidates, move applications, send messages, and schedule interviews, all under your company’s identity.

This is infrastructure. There is no built-in chatbot — you bring the orchestration.

What agents can do

• Read and create jobs — list open roles, create new postings, update details, delete (with confirmation)
• Rank and search candidates — access the scored, fairness-audited ranking list; run consent-aware candidate discovery
• Move applications — advance or reject candidates through pipeline stages
• Source candidates — express recruiter interest before a candidate applies
• Communicate — send messages to matched candidates, list conversation threads
• Schedule interviews — create interview events; candidates are notified automatically
• Trigger ranking refresh — re-run match scoring for a job and poll task status
• Read analytics — funnel stats, application counts, conversion rates

What agents cannot do

• Reveal candidate identity — identity reveal (name, email, contact details) stays a human action in the company portal. No tool exposes it.
• Write match config or pipeline definitions — scoring criteria and pipeline stages are set by humans.
• Bypass consent or anonymization — agents call the same backend endpoints as human recruiters; all existing gates apply.

Architecture in brief

Your LLM agent stack
        │  MCP Streamable HTTP — Authorization: Bearer jm_live_…
JobMatch gateway (port 8002)
  ├─ Key auth → scope check → rate limit
  ├─ Calls backend on your behalf (scoped 5-min JWT)
  └─ Re-strips identity fields before returning data
        │
JobMatch backend (port 8000)
  └─ All consent / anonymization / fairness gates apply unchanged

Agents never touch the database directly. Every call goes through the same permission layer human recruiters use — tenancy, anonymization, and fairness checks are structurally un-bypassable.

Quickstart

1. Create an agent

In the company portal, go to Settings → Agents → Create agent. Choose a name, description, and a role bundle that matches what the agent should do.

2. Create an API key

Under your agent, create an API key. The plaintext key (jm_live_… or jm_test_…) is shown exactly once — copy it now.

The key carries the scopes from your agent’s role bundle. You can also specify scopes manually to further restrict what this key can do.

3. Connect your MCP client

Point any MCP-capable client at the gateway:

Endpoint:  https://agents.job-match.cl/mcp
Transport: Streamable HTTP
Auth:      Authorization: Bearer jm_live_<your-key>

4. Self-configure with get_platform_context

Have your agent call get_platform_context first. It returns:

• Your company name and active jobs
• Available tools filtered to your key’s scopes
• Pipeline stages and their IDs
• Platform rules (what to do, what to never do)
• Role directives if your key is tied to a role bundle

Most agents only need this one call to bootstrap themselves.

5. REST fallback

If your client isn’t MCP-native, use the REST projection:

GET  /agent/v1/tools              # list tools for your key's scopes
POST /agent/v1/tools/{tool-name}  # body = the tool's arguments object

Same auth, same security pipeline. Identical results.

---

Next: Available tools · Role bundles · Webhooks