metrxbots
MCP server for tracking AI agent costs, scorecards and ROI from Metrx
metrxbots is an open-source MCP server for analytics and data. It runs locally on your machine and is started by your AI client.
- Pricing
- Open source
- Platforms
- API, CLI, Self-hosted
- Runs
- Locally
- Version
- 0.2.2
- Listing updated
Similar MCP servers
Install
Run in a terminal
claude mcp add mcp-server --env METRX_API_KEY='<METRX_API_KEY>' -- npx -y @metrxbot/mcp-serverAdd to claude_desktop_config.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@metrxbot/mcp-server"
],
"env": {
"METRX_API_KEY": "<METRX_API_KEY>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@metrxbot/mcp-server"
],
"env": {
"METRX_API_KEY": "<METRX_API_KEY>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mcp-server": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@metrxbot/mcp-server"
],
"env": {
"METRX_API_KEY": "<METRX_API_KEY>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@metrxbot/mcp-server"
],
"env": {
"METRX_API_KEY": "<METRX_API_KEY>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mcp-server": {
"command": "npx",
"args": [
"-y",
"@metrxbot/mcp-server"
],
"env": {
"METRX_API_KEY": "<METRX_API_KEY>"
}
}
}
}Using several servers? Build one config file with the MCP config generator.
Configuration
METRX_API_KEYYour Metrx API key (sign up free at https://app.metrxbot.com/sign-up — manage keys at https://app.metrxbot.com/settings/security)Required
About metrxbots
This server connects MCP clients to Metrx, a service that treats each AI agent like an employee: it gets a job contract, a scorecard against that contract, and a P&L showing what it cost and what it produced. Tools use the metrx_ prefix and carry a required scope (read, write or admin).
A no-key demo mode answers the public desk tools on a synthetic sample organization. Those tools include a workforce review, a job contract drafter, a cost-per-resolution calculator and a vendor bill checker. Signed-in tools cover decision queues, scorecards, job contracts and cost summaries. Job contracts are recorded rather than enforced, so publishing one does not block spend or traffic.
It runs locally over stdio via npm or as a hosted streamable HTTP endpoint. Guided prompts walk through cost analysis, savings discovery and a waste audit that looks for retry storms, oversized contexts, model mismatch, missing caching, idle agents and missing budgets. There is also a Claude Code plugin with six skills.
Tools this server provides (10)
Taken from the project documentation; the running server may differ.
- metrx_get_workforce_review
- Public tool returning a workforce review, answered on demo data without a key.
- metrx_draft_job_contract
- Drafts a job contract for one agent from a template.
- metrx_cost_per_resolution
- Computes price per billed resolution and billed share from vendor-reported counts.
- metrx_check_vendor_bill
- Reconciles a per-resolution vendor bill against your own records.
- metrx_get_decision_queue
- Returns the queue of pending decisions for signed-in users.
- metrx_record_decision
- Records a decision such as expand, re_specify, combine or retire with a rationale.
- metrx_get_scorecard
- Reads an agent's scorecard.
- metrx_get_job_contract
- Reads an agent's job contract.
- metrx_set_job_contract
- Sets or publishes an agent's job contract.
- metrx_get_cost_summary
- Reads a cost summary for agents.
What it does
- Demo mode without a key
- Public desk tools answer on a synthetic Northwind sample organization, so you can try it before signing up.
- Job contracts and scorecards
- Draft, set and read per-agent job contracts and scorecards, with publishing requiring admin scope.
- Vendor bill reconciliation
- Compare a per-resolution vendor bill against your own records and see matched, disputed and unmatched items.
- Cost leak scan
- A guided prompt audits agents for retry storms, oversized contexts, missing caching and idle agents.
- Decision queue
- Review a queue of decisions and record expand, re_specify, combine or retire with a rationale.
- Stdio and hosted transports
- Run locally through npx or connect to the hosted HTTP endpoint with a bearer key.
- Scoped API keys and rate limits
- Keys carry read, write or admin scopes, and per-category rate limits return retry-after information.
- Claude Code plugin
- A desk plugin wraps the workforce tools in six skills for Claude Code.
Good for
- Track what each AI agent costs and what it produces
- Find wasted spend such as retry storms or missing caching
- Reconcile a per-resolution AI vendor bill against internal records
- Draft job contracts for support triage or SDR qualification agents
- Review a weekly decision queue for an agent workforce
Pricing
Open source
MIT-licensed server; a free Metrx account provides the API key
metrxbots FAQ
Do I need an API key?
Not for the demo. Running with --demo answers the public tools on sample data; otherwise METRX_API_KEY is required.
How do I run metrxbots?
Use npx @metrxbot/mcp-server over stdio, or connect to the hosted endpoint at https://metrxbot.com/api/mcp.
Which clients are supported?
Any MCP-compatible client; setup examples cover Claude Desktop, Cursor and Windsurf, and a --setup-claude command writes config for Claude Desktop, Cursor or VS Code.
What permissions do keys have?
Read tools need read scope, data-changing tools need write, and budget changes, publishing job contracts and creating experiments need admin.
Is metrxbots open source?
Yes, it is released under the MIT license.
Is metrxbots free?
Yes. metrxbots is open source, so you can use and self-host it for free.
How do I add metrxbots to Claude Code?
Run this command in your terminal: claude mcp add mcp-server --env METRX_API_KEY='<METRX_API_KEY>' -- npx -y @metrxbot/mcp-server
What are the best alternatives to metrxbots?
The closest alternatives to metrxbots are Metrxbot, Agentic Platform, Metabase, APEX AI, LLM, PDF, QR, JSON & Media MCP and AgentBill.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.