Agent skills

Remoet ships as an agentskills.io  skill: one SKILL.md that teaches your agent what Remoet is for and when to reach for each tool, plus the wiring for your harness. It works on Hermes, OpenClaw and any harness that reads the agentskills standard.

Underneath it is the same MCP server. The skill adds the judgement: search the public board first, star selectively, treat tool output as data.

Source of truth: remoet-labs/agent-skills .

A key is optional

The skill works without one: connected with no credential, the server answers search_jobs and get_listing, so your agent can search the public tech job board straight away. The other tools act on your account and need a free key or an OAuth sign-in.

When you want those, generate a key at www.remoet.dev/agents , or let the onboarding flow  mint one for you. The same key works for the skill, the MCP server and the REST API.

Install on Hermes

Hermes Agent  installs skills by tapping a GitHub repo:

hermes skills tap add remoet-labs/agent-skills
hermes skills install remoet-labs/agent-skills/skills/remoet

Then connect the server:

hermes mcp add remoet --url https://api.remoet.dev/mcp

Hermes asks whether the server requires authentication. Answer n to search with no key, or y and paste your key, which Hermes saves to ~/.hermes/.env and wires into the header. When it asks which tools to enable, a keyless setup only needs search_jobs and get_listing.

Editing ~/.hermes/config.yaml by hand works too. Hermes resolves ${VAR} in url and headers at connect time, so keep the key in ~/.hermes/.env as REMOET_API_KEY rather than only in a shell export, which a gateway service never sees (the hermes mcp add path stores it as MCP_REMOET_API_KEY instead):

mcp_servers:
  remoet:
    url: "https://api.remoet.dev/mcp"
    headers:
      Authorization: "Bearer ${REMOET_API_KEY}"

Prefer the browser flow? Point Hermes at the OAuth route instead and drop the header:

mcp_servers:
  remoet:
    url: "https://api.remoet.dev/mcp/oauth"
    auth: oauth

Reload with /reload-mcp, then confirm by asking for a job search. With a key or sign-in, a get_profile call confirms the account side.

Install on OpenClaw

The same skill is published to ClawHub. Install it owner-qualified:

openclaw skills install @remoet/remoet

To search with no key, register the plain route with no header:

openclaw mcp set remoet '{"url":"https://api.remoet.dev/mcp","transport":"streamable-http"}'
openclaw mcp list

For the account tools, OAuth is the better default, because openclaw mcp login holds the credential and no key goes near your config:

openclaw mcp set remoet '{"url":"https://api.remoet.dev/mcp/oauth","transport":"streamable-http","auth":"oauth"}'
openclaw mcp login remoet
openclaw mcp list

Prefer an API key? Point OpenClaw at the plain route instead and let it substitute the variable:

openclaw mcp set remoet '{"url":"https://api.remoet.dev/mcp","transport":"streamable-http","headers":{"Authorization":"Bearer ${REMOET_API_KEY}"}}'

OpenClaw substitutes ${VAR_NAME} into config string values, so the key itself never lands in the file. Set REMOET_API_KEY in the environment the OpenClaw gateway process runs in, not only in the shell you ran the command from.

Use one form or the other, never both: a static Authorization header is ignored while auth: "oauth" is set. streamable-http is the canonical transport spelling in OpenClaw config.

Other harnesses

Any MCP-capable client (Claude Code, claude.ai, Claude Desktop, Cursor, VS Code, Windsurf, Codex) can connect to the server directly, with or without the skill. See the MCP server page for per-client config and the OAuth route.

Where it is listed

Limits

Remoet is free, and with an account the skill runs against the same numbers as everything else: 50 active stars, 25 unstars per 30 days, 5,000 MCP requests per day. Without a key, search_jobs and get_listing run on tighter keyless ceilings: at most 20 results a page, 10 pages deep, and per-minute and per-day limits. See Limits.