AgentLedger MCP server

AgentLedger — Per-Agent Spend Management

Track AI agent spend across x402/MPP/API-key rails, set budget caps, catch

spending anomalies, keep audit trails. The "Datadog for agent spending" —

the #1 verified community ask from agent builders.

🎯 Free tier: 3 agents per workspace, no card, no login. POST /start mints the workspace and shows the key once.

Connect AgentLedger

Connecting takes no auth handshake. Before your first write you need a workspacekey**, which you get one of two ways:

X-PAYMENT header. The paying wallet becomes the workspace identity — no email, no login, no human in the loop.

• Autonomous agent with a wallet: POST /v1/billing/x402 with an

/start — no signup, no login, no card. The workspace_key is shown once, right there, and the page carries the upgrade link for that workspace.

• Human: open

The key claims new agentids. Each claim mints that agent's own agentsecret, which is what authenticates every later write — the workspace_key is never needed again for that agent.

Claude Code:

claude mcp add --transport http agent-ledger https://aiagentscity.com/mcp/

Codex:

codex mcp add agent-ledger --url https://aiagentscity.com/mcp/

Cursor — merge into mcp.json:

{
  "mcpServers": {
    "agent-ledger": {
      "url": "https://aiagentscity.com/mcp/"
    }
  }
}

Any other MCP client: point it at the streamable-http remote https://aiagentscity.com/mcp/ — no headers, no auth handshake required to connect.

Try these prompts once connected:

• "Track my Claude Code spend"
• "Alert when any agent exceeds $50/day"
• "Weekly P&L report"

Try it — get a key, then track

# Track a spend — the FIRST call for a new agent_id CLAIMS it, so it needs
# your workspace_key, and it returns that agent's agent_secret.
# Save the secret: every later write (track/budget) to this agent_id uses it
# instead, and needs no workspace_key.
curl -X POST https://aiagentscity.com/v1/track \
  -H "Content-Type: application/json" \
  -H "AL-API-Version: 2026-09-01" \
  -d '{"agent_id":"my-agent","rail":"x402","amount_cents":100,"service":"search_query","workspace_key":"YOUR_WORKSPACE_KEY"}'

# Set a monthly budget cap — pass the agent_secret from above
curl -X POST https://aiagentscity.com/v1/budget \
  -H "Content-Type: application/json" \
  -H "AL-API-Version: 2026-09-01" \
  -d '{"agent_id":"my-agent","monthly_cents":5000,"agent_secret":"YOUR_SAVED_SECRET"}'

# Spend report + anomalies — requires the agent_secret from above
curl https://aiagentscity.com/v1/report/my-agent \
  -H "X-Agent-Secret: YOUR_SAVED_SECRET"

Want to just look at it? Every agent has a human-readable page (send the secret as a header, no curl needed for the JSON): https://aiagentscity.com/v1/report/my-agent/html

/v1/report, /v1/tokens, and /v1/alerts all require either X-Agent-Secret: or X-Workspace-Key: — missing or wrong credential gets 401. Headers only: there is no cookie or session arm, and no unauthenticated read path on REST or MCP. This is what stops a stranger from reading or overwriting someone else's agent_id.

Note this is a deliberate break from AgentLedger's earlier "share the link, anyone can read it" behavior — traded for real isolation between customers.

Two lines to a metered, capped agent

Shortened. The full README is on GitHub.