lamdis-plugins

A Claude Code plugin marketplace for the Lamdis exchange: a marketplace where AI agents pay people for physical work and get verified evidence back.

One plugin, lamdis, which does two things:

• connects the exchange's remote MCP server (https://exchange.lamdis.ai/mcp, streamable HTTP, no credential needed for the guest tools);
• ships a skill, lamdis, that the agent reads before using the tools: when to check feasibility, how the money works, what a receipt proves, and what never to tell the person.

Install

Two commands, in Claude Code's CLI:

claude plugin marketplace add lamdis-ai/lamdis-plugins
claude plugin install lamdis@lamdis-plugins

Or inside a Claude Code session:

/plugin marketplace add lamdis-ai/lamdis-plugins
/plugin install lamdis@lamdis-plugins

Restart Claude Code, or run /reload-plugins. claude mcp list should then show the lamdis server connected, and eight tools are available with no credential: check_feasible, observe_world, do_in_world, find_out, job_status, job_receipt, job_evidence, list_bids.

Try it with nothing

Coverage today is zero: nobody has registered as an operator yet, so a live check_feasible honestly answers reachable: "none" everywhere. The sandbox is how to see the whole loop run. Ask Claude to check feasibility with sandbox: true, then post an observation with sandbox: true; a simulated operator takes it, generated evidence is verified, and a receipt arrives in about ten seconds. Every reply about a sandbox job says in words that nobody was dispatched and nothing was paid.

A real job

A job posted with no credential comes back with a pay_at link and a token. The agent hands the person the pay link; their card is authorised for the job's ceiling and not charged. The job goes on the board when that lands, and the card is charged once, at the end, for exactly what was paid out on proof. The token is the only key to that job. The skill covers all of this.

Using an account

An agent key (lam_sk_…, issued from an account on the exchange) unlocks the full buyer side: projects, sites, vendors, open bidding, balance. On /mcp it goes in the Authorization: Bearer header. The plugin ships without one on purpose. To use a key, add the server yourself with the header, at user or project scope:

claude mcp add --transport http lamdis-account https://exchange.lamdis.ai/mcp \
  --header "Authorization: Bearer lam_sk_..."

Or copy plugins/lamdis/.mcp.json into your project as .mcp.json and add a headers block:

{
  "mcpServers": {
    "lamdis": {
      "type": "http",
      "url": "https://exchange.lamdis.ai/mcp",
      "headers": { "Authorization": "Bearer lam_sk_..." }
    }
  }
}

The skill applies either way. An agent key can spend and read what it bought; it cannot issue another key, raise a limit, connect a payout account or submit evidence.

Layout

.claude-plugin/marketplace.json     the marketplace
plugins/lamdis/
  .claude-plugin/plugin.json        the plugin manifest
  .mcp.json                         the remote MCP server
  skills/lamdis/SKILL.md            the operating manual the agent reads

Where the numbers come from

Every figure in the skill is taken from the exchange's source in lamdis-ai/lamdis-protocol: the fee (node/internal/exchange/settle.go), the worker ceilings (node/internal/api/assurance.go), the receipt confidence (node/internal/verify), the guest reply (node/internal/exchange/guest.go) and the tool definitions (node/internal/mcp/exchange.go). Where this repository and the running exchange disagree, the exchange is right.

License

Apache-2.0.