AgentLedger MCP¶
Remote MCP server (Streamable HTTP, agent-secret writes) - the "Datadog for agent spending": track what every autonomous agent spends across x402, MPP and API-key payment rails, cap budgets per agent, and keep a full audit trail. Five tools on one hosted endpoint, free during beta for up to 3 agents.
Server type: Remote (Streamable HTTP)
Auth: agent_secret issued on first track call (reads are open)
Endpoint: https://agent-ledger-production-0ff8.up.railway.app/mcp/
Tools: 5 (ledger_track, ledger_set_budget, ledger_report, ledger_alerts, ledger_list_agents)
Pricing: Free beta (up to 3 agents); Pro $19/mo unlimited
Category: Finance
Built by: entradox (independent software service), registry io.github.entradox/agent-ledger
Why This Matters for Operators¶
When agents start spending money autonomously - x402 micro-payments, MPP invoices, API keys handed to sub-agents - spend stops being a line item and becomes a stream of small, invisible transactions. AgentLedger gives that stream a ledger: every dollar and token an agent moves lands in one place, tagged by agent, rail and service.
The budget cap is the real control: a track() call that would cross a monthly cap is blocked with a 402, not just logged after the fact. And because every write requires the agent_secret issued on first registration, a rogue or compromised agent cannot silently inflate another agent's ledger.
Tools & Capabilities¶
| Tool | Purpose |
|---|---|
ledger_track |
Record a spend event (dollars and tokens) with agent_id, rail, service and model |
ledger_set_budget |
Set a monthly cap per agent; enforced at track time with a 402 |
ledger_report |
Pull a per-agent spend report |
ledger_alerts |
Anomaly alerts when spend spikes |
ledger_list_agents |
List all tracked agents |
A REST API mirrors the MCP surface: POST /v1/track, POST /v1/budget, GET /v1/report/{agent}, GET /v1/tokens/{agent}.
Installation¶
claude mcp add agent-ledger --transport http https://agent-ledger-production-0ff8.up.railway.app/mcp/
Configuration¶
{
"mcpServers": {
"agent-ledger": {
"type": "http",
"url": "https://agent-ledger-production-0ff8.up.railway.app/mcp/"
}
}
}
No signup and no API key: the first track call for a new agent_id returns an agent_secret in the response. Save it - every later write to that agent_id must include it. Reads stay open with no secret.
Business Relevance¶
- Operators running fleets of agents get one spend ledger across every payment rail instead of stitching together Stripe, x402 wallets and API-key invoices.
- Finance teams get a real audit trail - every dollar an autonomous agent moves is attributable to an agent, a rail and a service.
- Engineering leads get hard budget caps that fail closed (402) instead of post-hoc surprise invoices.
- Anyone evaluating agent ROI can compare per-agent spend against per-agent output with the token burn report.
Integration with CorpusIQ¶
AgentLedger pairs with CorpusIQ's own agent operations: while CorpusIQ runs the business - Stripe revenue, QuickBooks books, HubSpot pipeline - AgentLedger tracks the agent side of the cost equation. A composed workflow: CorpusIQ reports monthly platform spend, AgentLedger reports monthly agent-rail spend, and the two feed the same monthly operating review. Operators can ask one assistant "what did automation cost us this month across platform and agents" and get a complete answer instead of two separate exports.
Limitations¶
- Brand new - no track record yet; v0.1-era remote endpoint on a railway.app host.
- Beta free tier caps at 3 agents; real fleets need the $19/mo Pro plan.
- No self-host option published - the ledger lives on the vendor's endpoint.
- Write auth is a per-agent secret string, not scoped OAuth - treat the secret as a credential.