For reads, yes — headlessly: mint a read-only API token and your tool reads your instances, usage, deployments, skills, and billing status with an Authorization: Bearer header. For writes (create/patch/delete, connect a channel, rotate a key), the tool acts on your behalf inside your signed-in browser session; token-based writes are on the roadmap.
For developers & AI agents
Read-only API tokens for your agents and CI — plus AI-citable hosting
Mint a scoped, revocable token and read your Lobsterland instances, usage, and deployments headlessly. Answer engines can cite the product; your AI coding tools can read your account with a bearer token, or drive it on your behalf when you're signed in.
In one line
A signed-in user mints a read-only personal access token (shown once, stored as a hash) and sends it as Authorization: Bearer llp_… to read instances, usage, deployments, skills, and billing status. Writes still go through your signed-in session (token-based writes are on the roadmap). Lobsterland also publishes an llms.txt index, a text/markdown homepage mirror, an RFC 9727 API catalog, a filtered OpenAPI, and two agent-skills so assistants can discover and cite it.
Read via token
Scoped, revocable llp_… tokens for CI and AI tools.
Write on your behalf
State changes go through your signed-in session.
Discoverable
llms.txt, a filtered OpenAPI, and two agent-skills.
Discoverable by AI answer engines
Machine-readable, so assistants can cite you
- A curated llms.txt: a hand-picked index of the canonical product, setup, and security pages for AI crawlers to read first.
- A
text/markdownhomepage mirror: request the homepage withAccept: text/markdownand get a clean, parseable version of the product surface. - RFC 8288 discovery links in the page head: the api-catalog, agent-skills index, and llms.txt are advertised as link relations, so tools that read them — Cursor, Continue, GitHub Copilot, and answer engines — can find the surface.
Two published agent-skills
Skills your AI tools can install and use
Both are listed in the agent-skills index:
lobsterland-docs: a read-only skill that answers questions about the product and cites the public docs. It never touches a live instance.openclaw-setup: an operational skill that drives the dashboard — create instances, attach a provider, connect channels, manage skills, schedule crons — on behalf of an authenticated user.
The important caveat
The operational skill can read your account with a personal access token, but writes (create/patch/delete, connect a channel, rotate a key) require your signed-in browser session. A read token is rejected on write routes — token-based writes are on the roadmap, not in this release.
How authentication actually works
Read with a token, write with your session
Reads use a personal access token. You mint one in the dashboard under Developer / API tokens: pick a name, read scopes (instances:read, usage:read, skills:read, workspace:read, billing:read), and an expiry. The secret (llp_…) is shown once and stored only as a hash, and you can revoke it any time. Send it as Authorization: Bearer llp_…. A token is accepted only on scoped GET routes, and responses have secret-looking fields redacted.
Writes — anything that changes state — are authenticated by your own signed-in session (an HTTP-only cookie). A read token is rejected on write routes by design; an AI agent performs writes on your behalf inside your authenticated session.
That's the honest boundary today: read your account headlessly with a scoped token, and write on your behalf when you're signed in. Token-based writes and an OAuth agent-registration flow are on the roadmap; there is no anonymous or third-party token path.
What an agent can do for you
The same actions the dashboard exposes
Acting on your behalf, an assistant can drive the operations you'd otherwise click through yourself:
Instances
Create and manage hosted OpenClaw or Hermes instances.
Providers
Attach a model provider by API key or OAuth.
Channels
Connect Telegram, Slack, or WhatsApp.
Skills & cron
Manage skills and schedule cron jobs.
The read surface is described by a filtered OpenAPI document that advertises the bearer-token scheme, so your coding tool can learn exactly what a token can call. Writes are driven through the dashboard. See all features.
FAQ
Developer and agent questions
Yes — read-only personal access tokens. A signed-in user mints a scoped, revocable token (Developer / API tokens), shown once and stored only as a hash. It authenticates GET requests via Authorization: Bearer llp_… over scopes like instances:read and usage:read. Writes still require your signed-in session; there are no anonymous or third-party tokens.
Two: lobsterland-docs (read-only product Q&A that cites the public docs) and openclaw-setup (drives the dashboard on behalf of an authenticated user). They're listed in /.well-known/agent-skills/index.json.
Two ways: for read-only access you mint a personal access token and the tool sends it as Authorization: Bearer llp_…; for anything that changes state it acts inside your own browser sign-in (Google or email), which establishes a session cookie. A read token is rejected on write routes by design.
Yes — a curated /llms.txt index and a text/markdown homepage mirror for answer engines, plus a machine-readable OpenAPI description of the dashboard API published for discovery.
Ready
Read your account with a token. Drive it when you're signed in.
Scoped, revocable read tokens for CI and AI tools. Discoverable by answer engines. Writes on your behalf, in your session.