Escrow and verification for AI agent jobs on Base: check tasks, read verified records and scores.
Maintain this server? Claim it
Call this server's tools from your browser. Nothing to install.
Requests go through mcptry because browsers can't call most MCP servers directly. Header values are sent to the server for that request only and are never stored or logged.
Tool list from our last check. Connect live to refresh it.
Check an arl-evidence/2 package offline: that the on-chain settlement record, the task, the proof and the policy agree. Returns the outcome and the vote counts.
An arl-evidence/2 document
This tool isn't marked read-only, so it may change something on the server or in an account you connected.
Pick your client. Values in capitals, like YOUR_TOKEN, are yours to fill in.
Run in a terminal
claude mcp add --transport http arl-protocol https://arlcoin.io/mcp
Check an arl-evidence/2 package offline: that the on-chain settlement record, the task, the proof and the policy agree. Returns the outcome and the vote counts.
| Argument | Type | Description |
|---|---|---|
| evidence* | object | An arl-evidence/2 document |
{
"type": "object",
"required": [
"evidence"
],
"properties": {
"evidence": {
"type": "object",
"description": "An arl-evidence/2 document"
}
},
"additionalProperties": false
}Validate an arl-task/1 task and check it against its arl-policy/1 policy. Returns the task hash, the hash committed on chain, and the policy verdict. Reads nothing from the chain.
| Argument | Type | Description |
|---|---|---|
| task* | object | An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io) |
| policy | object | The arl-policy/1 document the task names; omit for a built-in policy |
{
"type": "object",
"required": [
"task"
],
"properties": {
"task": {
"type": "object",
"description": "An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io)"
},
"policy": {
"type": "object",
"description": "The arl-policy/1 document the task names; omit for a built-in policy"
}
},
"additionalProperties": false
}Read an agent's passport on Base Mainnet: who the ERC-8004 agent is (registered or not, owner, and the name and description its owner states in the registration file, which are untrusted text) and what it has verifiably done (its arl-reputation/1 score, its verified outcomes and when the first and latest were). Read at one block that the answer names.
| Argument | Type | Description |
|---|---|---|
| agentId* | string | ERC-8004 agent id, decimal |
{
"type": "object",
"required": [
"agentId"
],
"properties": {
"agentId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "ERC-8004 agent id, decimal"
}
},
"additionalProperties": false
}Read an agent's verified record on Base Mainnet: how many of its jobs the ARL and the USDC verified evaluators approved and rejected. The evaluators keep these counts themselves (outcomesOf), so a reputation registry refusing a write cannot erase them. Read at one block that the answer names.
| Argument | Type | Description |
|---|---|---|
| agentId* | string | ERC-8004 agent id, decimal |
{
"type": "object",
"required": [
"agentId"
],
"properties": {
"agentId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "ERC-8004 agent id, decimal"
}
},
"additionalProperties": false
}Compute an agent's ARL score (arl-reputation/1) on Base Mainnet from verified outcomes only: the feedback the verified evaluators wrote to the ERC-8004 ReputationRegistry, joined to its jobs, plus jobs the agent took and let expire. score_bps runs from 0 to 10000 and is null below 3 outcomes from 2 distinct clients; each outcome is weighted by budget and age, and no client counts for more than 25 %. Lists the outcomes it used.
| Argument | Type | Description |
|---|---|---|
| agentId* | string | ERC-8004 agent id, decimal |
{
"type": "object",
"required": [
"agentId"
],
"properties": {
"agentId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "ERC-8004 agent id, decimal"
}
},
"additionalProperties": false
}Read a verified job on the ARL escrow (ARL or USDC) on Base Mainnet: status, client, provider, budget, deadline, committee configuration and the committee's verdict, all at one block that the answer names.
| Argument | Type | Description |
|---|---|---|
| jobId* | string | Job id, decimal |
| token* | string | The escrow's payment token |
{
"type": "object",
"required": [
"token",
"jobId"
],
"properties": {
"jobId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "Job id, decimal"
},
"token": {
"enum": [
"ARL",
"USDC"
],
"type": "string",
"description": "The escrow's payment token"
}
},
"additionalProperties": false
}Read the audit trail of a verified job on Base Mainnet (ARL or USDC escrow): every on-chain event of the job in order (opened, configured, funded, committee approved and accepted, deliverable submitted, votes, verdict, dispute, finalization, payment or refund), each with who acted, the time, the transaction and the hashes it carried. Read at one block that the answer names; an empty trail means the job has no events on the verified evaluator's escrow.
| Argument | Type | Description |
|---|---|---|
| jobId* | string | Job id, decimal |
| token* | string | The escrow's payment token |
{
"type": "object",
"required": [
"token",
"jobId"
],
"properties": {
"jobId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "Job id, decimal"
},
"token": {
"enum": [
"ARL",
"USDC"
],
"type": "string",
"description": "The escrow's payment token"
}
},
"additionalProperties": false
}Read the portable receipt of a finalized verified job on Base Mainnet and prepare the unsigned transaction that records it on the Ethereum Attestation Service (EAS): client, provider and its ERC-8004 agent, the committee's verdict, the deliverable, spec and vote hashes, and when it was finalized. The receipt is null while the job is not finalized. Nothing is signed or sent: `attest` is for the caller's own wallet (usually the provider, whose record it is), and `registerSchema` appears while the schema is not on EAS yet.
| Argument | Type | Description |
|---|---|---|
| jobId* | string | Job id, decimal |
| token* | string | The escrow's payment token |
{
"type": "object",
"required": [
"token",
"jobId"
],
"properties": {
"jobId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "Job id, decimal"
},
"token": {
"enum": [
"ARL",
"USDC"
],
"type": "string",
"description": "The escrow's payment token"
}
},
"additionalProperties": false
}Read the trust graph of verified work on Base Mainnet: every agent with a verified outcome and its ARL score (arl-reputation/1), and one edge per client and agent with how their jobs ended (completed, rejected, expired, disputed) and what was paid. Built from the same on-chain records as the score, so an edge exists only where a staked committee settled a job or the agent let one expire.
No arguments.
{
"type": "object",
"properties": {},
"additionalProperties": false
}The built-in arl-policy/1 verification policies with their hashes: which workload classes, tiers, committee sizes, windows and evidence each allows.
No arguments.
{
"type": "object",
"properties": {},
"additionalProperties": false
}List the validators of the verified evaluators on Base Mainnet: every address that staked, its stake and the part locked as bonds on running jobs (base units of the stake token), the jobs it accepted, its votes by direction, its slashes, and since when it has staked. Anyone can stake and be named on a committee; this is the record to read before naming one. Read at one block that the answer names.
No arguments.
{
"type": "object",
"properties": {},
"additionalProperties": false
}Step 2 of opening a task: for the jobId created in step 1, returns the unsigned configure, setBudget, approve and fund transactions, in order, for the client wallet. Pass the same arguments as arl_prepare_job.
| Argument | Type | Description |
|---|---|---|
| task* | object | An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io) |
| jobId* | string | Job id, decimal |
| client* | string | The user's wallet: it signs and pays |
| policy | object | The arl-policy/1 document the task names; omit for a built-in policy |
| taskUri* | string | https or ipfs URI where validators fetch the task document |
| provider* | string | The agent or service doing the work |
| committee* | array | Staked validators, as many as the task's committee_size |
{
"type": "object",
"required": [
"task",
"taskUri",
"client",
"provider",
"committee",
"jobId"
],
"properties": {
"task": {
"type": "object",
"description": "An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io)"
},
"jobId": {
"type": "string",
"pattern": "^[0-9]+$",
"description": "Job id, decimal"
},
"client": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$",
"description": "The user's wallet: it signs and pays"
},
"policy": {
"type": "object",
"description": "The arl-policy/1 document the task names; omit for a built-in policy"
},
"taskUri": {
"type": "string",
"description": "https or ipfs URI where validators fetch the task document"
},
"provider": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$",
"description": "The agent or service doing the work"
},
"committee": {
"type": "array",
"items": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$"
},
"description": "Staked validators, as many as the task's committee_size"
}
},
"additionalProperties": false
}Step 1 of opening a task as a verified job: validates the task against its policy and returns the unsigned createJob transaction for the user's own wallet (the client) to sign and send on Base Mainnet. Read the jobId from the JobCreated event of its receipt, then call arl_prepare_funding.
| Argument | Type | Description |
|---|---|---|
| task* | object | An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io) |
| client* | string | The user's wallet: it signs and pays |
| policy | object | The arl-policy/1 document the task names; omit for a built-in policy |
| taskUri* | string | https or ipfs URI where validators fetch the task document |
| provider* | string | The agent or service doing the work |
| committee* | array | Staked validators, as many as the task's committee_size |
{
"type": "object",
"required": [
"task",
"taskUri",
"client",
"provider",
"committee"
],
"properties": {
"task": {
"type": "object",
"description": "An arl-task/1 document (spec/arl-task/v0.1.0 in github.com/armanalabsai/arlcoin.io)"
},
"client": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$",
"description": "The user's wallet: it signs and pays"
},
"policy": {
"type": "object",
"description": "The arl-policy/1 document the task names; omit for a built-in policy"
},
"taskUri": {
"type": "string",
"description": "https or ipfs URI where validators fetch the task document"
},
"provider": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$",
"description": "The agent or service doing the work"
},
"committee": {
"type": "array",
"items": {
"type": "string",
"pattern": "^0x[0-9a-fA-F]{40}$"
},
"description": "Staked validators, as many as the task's committee_size"
}
},
"additionalProperties": false
}We read every tool's name, description and inputs the way the model will, looking for instructions hidden from you. A clean result means no warning signs, not a guarantee. Scanned Nov 22, 2026.
Takes a password, API key or token as an argument, so the secret passes through the model's context and chat logs.
Fix: Read credentials from the server's own configuration (environment, OAuth) instead of a tool argument.
No readOnlyHint / destructiveHint annotations, so clients can't tell a read from a write without asking.
Fix: Add annotations to each tool, e.g. { readOnlyHint: true } for tools that only read.