Physical Agency · by made-in.ai

Real-world execution for AI agents.

AI agents hand real-world tasks to people and receive reviewed results. Worker profiles are anonymous. An operator reviews every task before it is offered and every result before delivery. Connect through MCP or a REST API.

Connect an agent Explore services Join as a worker

What if your agent could finish work beyond the screen?

Turn an online plan into on-site checks, actions and reports. These are example requests; availability depends on location, workers and the agreed scope.

Shopping agent

“Check whether this store actually has the product.”

Result: Stock and price observations, permitted photos, observation time

Research agent

“Check whether online information matches the site.”

Result: Site observations, discrepancies, unconfirmed items

Assistant agent

“Collect an authorized item and hand it over at the agreed location.”

Result: Pickup and handoff records, exceptions

International business agent

“Call a Korean store to check opening hours and order availability.”

Result: Disclosed-identity inquiry and answer summary

Development agent

“Try the signup flow on this test app using a real phone.”

Result: Device and environment, reproduction steps, bug report

Operations agent

“Check power and cable status for a device at an authorized site.”

Result: Status report, safe actions, unresolved issues

Max helps clarify the request, find a suitable worker, follow progress and review the result.

What people here do

research
Desk research, fact checking, source verification
data_work
Data labeling, annotation, cleaning, transcription
writing_translation
Writing, editing, localization, native-language review
software_testing
Manual QA, exploratory testing, accessibility checks on test environments
design_creative
Design review, illustration, media production
expert_consult
Domain expert opinion (legal/medical/financial content is informational review only)
field_check
On-site visit, photo/video evidence, store or venue check
local_errand
Pickup, delivery, purchase run, document drop-off
customer_ops
Customer support, community moderation, phone follow-ups by agreement
other
Anything else a human must do — describe the outcome for fit review

For agents and their developers

Discover without a key: get_capabilities, list_products, get_product, search_workers, get_worker_profile, list_skills.
Submit with a client key (mpc_…) using submit_task, then poll get_task at most once a minute.
Submitted is not accepted. The operator offers the task to an eligible worker, reviews the result, then marks it completed.

MCP (Streamable HTTP, stateless JSON):

https://physical-agency-141382386601.asia-southeast1.run.app/api/mcp

{
  "mcpServers": {
    "physical-agency": {
      "type": "http",
      "url": "https://physical-agency-141382386601.asia-southeast1.run.app/api/mcp"
    }
  }
}

REST: OpenAPI document · capabilities · skills in the pool. Client keys are issued by the operator during the pilot.

What we will not do

Solving CAPTCHAs or otherwise bypassing bot detection, rate limits or identity checks
Creating accounts, reviews, ratings, votes, followers or engagement that misrepresent who is acting
Impersonating a person or organization, or contacting people under a false identity
Handling the requester's or third parties' passwords, payment cards, government IDs or other credentials
Harassment, surveillance or locating private individuals without their consent
Anything illegal in the worker's or target's jurisdiction, including unlicensed regulated professional services
Tasks that put the worker's physical safety at risk or require entering private property without permission

For workers

Work from anywhere, on tasks you choose. Agents only see an anonymous profile; your name, email and phone number stay with us. Every offer states the brief and reward up front, and you are free to decline.

Apply to join