MCP server

@ripar/skills is an MCP server over the live Algorand registries, plus x402 quoting and calling. Six of its tools read the chain, one composes an unsigned transaction, and one forwards a payment header you supply.

Connect

Not published to npm. Run it straight from the repository:

{
  "mcpServers": {
    "ripar": {
      "command": "npx",
      "args": ["-y", "github:nickthelegend/ripar-skills"],
      "env": { "RIPAR_NETWORK": "testnet" }
    }
  }
}

Or from a local clone, which is faster after the first build:

git clone https://github.com/nickthelegend/ripar-skills && cd ripar-skills
npm install && npm run build

{
  "mcpServers": {
    "ripar": {
      "command": "node",
      "args": ["/absolute/path/to/ripar-skills/dist/bin/mcp-stdio.js"],
      "env": { "RIPAR_NETWORK": "testnet" }
    }
  }
}

# Claude Code
claude mcp add ripar -- npx -y github:nickthelegend/ripar-skills

RIPAR_NETWORK is the only variable that matters, and testnet is the default — TestNet is the only network the registries are deployed on today. There is no token and no server-side spend cap to configure: see Security for where budget enforcement actually lives.

The registries it reads

Tools

ripar_search_agents

List or search agents in the IdentityRegistry by domain substring, exact agent id, or Algorand address. Read-only.

{ "query": "ripar.io", "limit": 25, "withReputation": true }

Returns live registry records and an /.well-known/agent.json URL for each. If the chain is unreachable it fails rather than guessing.

ripar_get_agent

One agent, by agentId, domain or address, optionally with its reputation score and its jobs. Read-only.

Returns found: false with a reason when the registry has no such agent — the contract's "not found" value is a literal 0, so an absent agent is a real answer, not an error.

ripar_get_reputation

An agent's score from the ReputationRegistry: payments credited, total USDC volume, and validator verdicts. Read-only.

ripar_list_jobs

Jobs on the ValidationRegistry, newest first, filterable by status, agentId or jobId. Read-only. Each job commits to its spec by hash; the spec and result themselves stay offchain.

{ "status": "submitted", "limit": 25 }

Statuses are open, assigned, submitted, validated, disputed, cancelled.

ripar_settlements

Real USDC transfers for an agent from the Algorand indexer, each marked with whether the ReputationRegistry has already counted it. Read-only.

Inbound payments marked counted: false are reputation the agent earned but was never credited for — a gap only visible because the transfer log and the registry are read together.

ripar_quote_endpoint

Ask a paid endpoint what it charges, without paying. Makes the request, reads the 402 challenge, and reports the cheapest acceptable payment: amount, asset, network, payee. Read-only.

{ "url": "https://api.ripar.io/api/summarize", "method": "POST", "body": { "text": "…" } }

An endpoint that answers 200 is reported as free rather than as an error.

ripar_call_endpoint

Call an endpoint and return its response. Can spend money, but only with a payment header you supply.

{
  "url": "https://api.ripar.io/api/summarize",
  "method": "POST",
  "body": { "text": "…" },
  "paymentHeader": "<base64 signed payment>"
}

Without paymentHeader, a 402 comes back as the challenge rather than a result. The header is forwarded untouched as X-PAYMENT.

ripar_post_job

Compose a ValidationRegistry post_job call and return it unsigned, as base64 msgpack, with a plain-language summary of what signing it would do. Nothing is submitted and no key is used.

{
  "sender": "KBDRZK3B…KEISKQ",
  "specHash": "<hex of the 32-byte sha256 of the spec>",
  "budgetMicro": 2500000,
  "validatorAgentId": 0
}

budgetMicro is USDC base units — 2500000 is $2.50. See Jobs & validation for what that number does and, importantly, does not do.

ripar_fund_job

Composes the two-transaction group that moves a budget into escrow: the asset transfer to the ValidationRegistry's own account, then the fund_job call that reads the amount off it. Grouped in that order, returned unsigned — the server holds no key.

Refuses before a fee is spent when the sender is not the client, the job is past funding, or the amount is zero.

ripar_settle_escrow

Composes release_escrow or refund_escrow, chosen with action.

Release needs a passing verdict, and may be called by the client — or by anyone once the dispute window has passed, because a validator who never returns would otherwise freeze the worker's money. Refund needs a failed verdict or a cancelled job, and pays the client whoever triggers it: the destination is read off the job, so calling it cannot redirect it.

Resources

Registry state, readable without calling a tool:

Both templates ship list callbacks that read the chain, so resources/list returns the real roster rather than URI patterns.

Prompts

Each references only tools that exist — asserted in the test suite, because a prompt naming a missing tool fails at the point somebody trusts it.

A full exchange

user  → Which registered Ripar agents have actually been paid?
tool  → ripar_search_agents({ query: "", withReputation: true })
tool  → 2 agents; agent 2 has a score box, agent 1 does not
tool  → ripar_settlements({ agentId: 2 })
tool  → 4 inbound USDC transfers, 3 counted, 1 counted:false
agent → Agent 2 has been paid four times but only three of those were
        credited to its on-chain score. Agent 1 has never been credited.

Every number there came off the chain. Nothing in this server fabricates one — a failed read is an error, never a plausible-looking zero.