One install
Everything your agent can do with verified humans
Hire
Hire human employees on demand
Your agent sources identity-verified experts for real work: code review, research, judgment calls, physical-evidence tasks. They are matched, screened, and verified on long-horizon jobs, and you pay only on completion.
200+ specialized domains, paid on verified completion
Train
Train and evaluate your models
Launch preference, rating, and red-teaming tasks mid-run. RLHF, reward modeling, and release-gating evals get verified human signal back as structured JSON, ready for your training loop.
Structured results with verification provenance
Listen
Hear from real customers
Recruit verified users and run AI-moderated interviews and surveys without leaving the workflow. When your agent hits ambiguity, it escalates to a human and feeds the answer back into context.
Verified humans, not crowdworkers or synthetic profiles
Tools, not API code
Your agent calls humans the same way it calls any tool
No SDK, no wrapper functions, no context switch. Tools are discovered automatically and your agent decides when to reach for a human, scope the work, check the quote, and read back verified results.
Agent session
# agent reasons, then acts
call terac_create_quote
{ "role": "Senior React engineer", "task": "review PR for race conditions", "count": 3 }
# quote returns before any spend
{ "quote_id": "qt_8f2", "price_usd": 84, "eta_hours": 6 }
call terac_launch_opportunity
{ "quote_id": "qt_8f2" }
# later: verified results, pay-on-completion
call terac_get_submissions
{ "status": "APPROVED", "expert_attestations": ["work_email", "gov_id"] }
How it works
From one install to verified human work
The missing step in your pipeline was always the human one. Now it is one tool call away.
Install the MCP server
One command in Claude Code, Claude Desktop, Cursor, or any MCP client. Tools are discovered automatically. No SDK, no wrapper code.
Your agent defines the job
In natural language, your agent describes who it needs and what they should do. Terac translates that into expert targeting, screeners, and attestations, then returns a quote before any spend.
Verified humans deliver
Vetted experts complete the work while our moderator screens and verifies. Structured, validated results return to your agent's context. You pay only for submissions that pass review.
Get started
Start where your stack already lives
MCP server
Works with any MCP client
Claude Code, Claude Desktop, Cursor, or any client that implements the Model Context Protocol. Tools are discovered automatically. Your agent decides when to call.
Install the MCP serverREST API
Full programmatic control
Opportunities, quotes, submissions, attestations, and webhooks over a typed REST API. The right fit for custom integrations and training pipelines.
Read the API docsPipelines & CI
Scriptable as a release gate
Launch a job, wait on completion, aggregate results against a threshold, pass or fail the run. Idempotent launches make retries safe from automated callers.
Talk to the teamFor AI teams
Put a human in the loop, in the pipeline
Connect your agent to verified expert labor in one install. Training data, evals, and customer signal come back as structured JSON your pipeline can act on.
For agent builders
The supply layer for human labor
When your agent needs the exact human for a task, it sources one on demand. Define the job, Terac gets it done, and you pay out only on verified completion.
FAQ
Question not answered? Email us at support@terac.com or book a meeting with our team.
Book a meetingYes. Install it in Claude Code, Claude Desktop, Cursor, or any MCP-compatible client. Your agent can scope a job, launch it to verified experts, and receive structured results without writing any API code. Tools are discovered automatically, so there is no SDK or wrapper to maintain.
Three things, from one install: hire vetted human experts for a task, collect human data to train and evaluate models, and gather feedback from real verified customers. Each is exposed as a tool your agent calls on its own when it hits work that needs a human.
Every expert is identity-verified with attestations (government ID, work email, LinkedIn) and continuously evaluated. Our AI moderator screens, interviews, and verifies on long-horizon tasks, catching effort-faking and fraud before results reach you. You are not getting anonymous crowdworkers or AI-generated profiles.
You pay on verified completion. Submissions that fail review do not bill. Quotes are returned before a job launches, so your agent can check cost against a budget and decide whether to proceed. Idempotent launches make retries safe from automated callers.
Yes. Submissions return as structured JSON with provenance on who completed each task and how they were verified. That feeds directly into RLHF, preference tuning, reward modeling, and release-gating evals, with no transformation step between the response and your run.
The MCP server works with any client that implements the Model Context Protocol. The same capabilities are available over our REST API for custom integrations and CI pipelines. Pick whichever fits where your stack already lives.
