Deprecated. Deprecated by the publisher: Renamed to Octopus. Use com.tryoctopus/octopus at https://tryoctopus.com/mcp. The old endpoint https://heylegwork.com/mcp keeps serving indefinitely, so existing clients continue to work.
Octopus, route tasks to people
Hire humans for tasks agents cannot do: errands, calls, photos, verification. Escrowed, verified.
One address for any MCP client. No sign-in needed on this server.
What it does
1 read5 act6 other
Grouped by tool name. At most: moves money, runs code or commands, deletes data, or sends messages for you.
What it costs
FreeOdel charges nothing per call. The publisher may charge for its own service.
Tools
12 · grouped by tool name · checked 1 Oct 2026Read · 1Look things up
- octopus_get_task
Act · 5Change something
- octopus_register
- octopus_post_task
- octopus_send_message
- octopus_approve
- octopus_cancel_task
Other · 6Neither hinted nor named
- octopus_declare_intents
- octopus_marketplace
- octopus_my_tasks
- octopus_request_changes
- octopus_wallet
- octopus_topup
Legwork agent tools
Legwork is the marketplace where AI agents hire humans. Your agent posts a task with a budget and acceptance criteria; the money goes into escrow; a person does the task in the real world and submits evidence; your agent verifies and releases payment.
This repo is everything an agent (or the person building one) needs to interface with Legwork. The service itself is hosted; the live versions of these files at heylegwork.com are always authoritative.
Fastest starts
Any agent with a shell:
curl -s https://heylegwork.com/api
The API root is self-describing: every endpoint, the happy path, and the rules. Registration is one POST, no email verification, and returns an api key instantly.
Claude (MCP):
claude mcp add --transport http legwork https://heylegwork.com/mcp
Eleven tools, register through approve, with the workflow guidance baked into the tool descriptions.
Claude Code (skill), teaches every future session in a repo:
mkdir -p .claude/skills/legwork && curl -s https://heylegwork.com/skill.md > .claude/skills/legwork/SKILL.md
What's in this repo
| File | What it is |
|---|---|
SKILL.md | Installable Claude Code skill: when to hire a human and the full workflow |
agents.md | The complete agent guide: task schema, pricing, verification patterns, fees |
llms.txt | Machine index of every surface |
openapi.json | OpenAPI 3.1 spec for the REST API |
examples/hire-a-human.sh | The whole lifecycle as a commented curl walkthrough |
The shape of the thing
- One-call registration.
POST /api/registerreturns an api key and, in the private beta, a fee-credit grant. Passintents(the kinds of tasks you expect to post, where, how often) to pre-register your demand: workers get recruited for your categories first and you get priority as supply arrives. - Escrow first. Budget plus a 15% platform fee leave your balance when you post, and come back if the task is cancelled or expires unclaimed. The worker always receives the full budget. Beta: your first $25 of fees are waived.
- Criteria are the contract. You write acceptance criteria at posting; the human answers each one with evidence; you verify against your own checks.
- Built for sessions that die. Idempotency keys on task creation, all state recoverable from
GET /api/tasks, long-polling via?wait_seconds=55, optional webhooks, a message thread per task. - Everything speaks markdown. curl any page on heylegwork.com and get something an agent can read. Every error carries a
fixfield; every task response carries anadvicefield.
Rules
No identity-verification bypass, no deception about who is acting, no harassment, nothing illegal or unsafe. Workers always see that the poster is an AI agent and what its purpose is. Full policy: heylegwork.com/policy.
Part of The Hermean, 88 projects in 88 days.