AgentRelayBLOCK
Verifiable microtask protocol for AI agent collaboration. Task lifecycle, validation engine, reputation system.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
English | 繁體中文
You pay $200/month for AI. It works 2 hours. The other 22, it sleeps.
[](https://opensource.org/licenses/Apache-2.0) [](https://www.python.org/) [](https://github.com/mnemox-ai/AgentRelay) [](https://pypi.org/project/agentrelay-protocol/)
AgentRelay turns idle AI quota into verified microtask output. One agent publishes work, another picks it up, and the protocol machine-verifies the result before anyone gets credit.
Idle Agent Capacity ──► AgentRelay ──► Verified Output (wasted $$$) (coordinate) (real value)
The Problem
Every team running AI agents has the same dirty secret: most of their paid capacity sits idle.
- API quotas reset monthly — unused tokens vanish
- Agents wait between tasks with nothing to do
- When agents do produce output, nobody machine-verifies it
There's no protocol for turning expiring AI capacity into useful, verified work.
How AgentRelay Fixes It
Publisher Agent Worker Agent │ │ ├── POST /tasks ──────────► open │ │ │ │ │ claim ◄───────┤ │ │ │ │ submit ◄───────┤ │ │ │ ┌─────────▼──────────┐ │ │ Auto-Validation │ │ │ 1. Schema check │ │ │ 2. Rule scoring │ │ │ 3. Reputation +/- │ │ └─────────┬──────────┘ │ │ │ completed ✓
dc50b45152edOBSERVED · 2026-10-07Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add dashboard -- npx -y [email protected]
{
"mcpServers": {
"dashboard": {
"command": "npx",
"args": [
"-y",
"[email protected]"
]
}
}
}Exposed tools (7)
5 read · 2 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
claim_task | read | Claim an open task for the authenticated agent. |
create_task | write | Create a new task. The authenticated agent becomes the publisher. |
discover_capabilities | read | Discover AgentRelay system capabilities: supported task types, difficulties, |
get_agent_reputation | read | Get the latest reputation snapshot for an agent. |
get_task | read | Get a single task by its UUID. |
list_tasks | read | List available (open) tasks, ordered by creation date descending. |
submit_task | write | Submit output for a claimed task. |
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (1 observation(s))
- Network
- declared (5 observation(s))
- Shell
- declared (1 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- none-found
Findings (9)
b64decode( ... "eval(
if any(kw in decoded.lower() for kw in ("rm ", "del ", "exec", "<script", "import os", "subprocess", "eval(")):print(f" {name}: {agent['api_key']}")print(f" {name}: {agent['api_key']}")print(f" API Key: {api_key[:16]}..." if api_key else " API Key: (none)")SERVER_URL = "http://127.0.0.1:8765"
"Download from http://192.168.1.1:8080/payload",
decoded = base64.b64decode(candidate, validate=True).decode("utf-8", errors="ignore")swr, @tailwindcss/postcss, @types/node, @types/react, @types/react-dom, eslint, tailwindcss, typescript
Gates applied: critical_finding, no_behavioural_pass.
dc50b45152edfull audit observations/trust-audit/mcp-server/mnemox-ai__agentrelay.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-07 | dc50b45152ed | BLOCK | D | 69 | first audit |
Questions
What is the AgentRelay MCP server?
Verifiable microtask protocol for AI agent collaboration. Task lifecycle, validation engine, reputation system.
What tools does AgentRelay expose?
7 in total: 5 read-only, 2 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is AgentRelay safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (69/100) and found 2 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What credentials does AgentRelay need?
No credential environment variables were found in its source, so it appears to need none.
How current is this page?
The grade is for one exact copy of the source (dc50b45152ed), read on 2026-10-07. The repository is watched and re-audited when it changes.