callx402 MCP server

Payload — Developer infrastructure for x402, agent payments, and programmable revenue. PAYLOAD → VEYLINE (flagship) → CALLX402 (action layer) → REVRULE (separate) → developer products → free utilities. This repo: callx402 by Payload — the universal action layer into Veyline's x402 infrastructure.

[callx402 logo]

[License: MIT] [Node >= 18] [MCP stdio]

Powered by Veyline. When x402 breaks, callx402.

callx402 is the universal action layer into Veyline's x402 infrastructure. Say what you need done in plain language (or hit an HTTP endpoint) and callx402 routes it to the production system that does the real work: diagnose a broken x402 payment, rescue a failed transaction, route an agent job to the cheapest viable path, resolve settlement state from on-chain evidence, or execute under an explicit budget with fail-closed safety rules.

Why it exists: x402 failures are expensive and opaque. A settlement attempt ends in settlement_pending and nobody knows whether the money moved. A 402 response your wallet misreads. A retry that signs a second authorization for the same intent and pays twice. callx402 exists for exactly those moments: diagnose pins the failure to a stage, rescue triages the incident, resolve settles the question from evidence, route finds the cheapest viable path, execute runs under a hard budget.

Try it (two minutes, no setup, no account):

npm install -g callx402callx402 statuscallx402 'complete this job for under $1'

status prints an honest per-subsystem reachability report, no credentials needed. The intent line parses and plans your request, then stops before executing anything: with no live tool endpoint wired it reports "no executable route available", so no money moves. (Use single quotes: in double quotes your shell would eat the $1.)

Free read-only checks against the live Payload rail:

curl https://payload-rail.fly.dev/v1/callx402/actionscurl 'https://payload-rail.fly.dev/v1/callx402/quote?action=diagnose'

The first returns the paid action catalog with prices; the second returns a free price quote for one action. Invoking a rail action is paid per action via checkout (see https://payloadhq.github.io/agents.json); these commands never pay anything.

If it saves you one debugging session, star the repo and read on.

MCP server: connect in 60 seconds

This repo ships a zero-dependency, read-only MCP server (mcp/index.js, node stdlib only, stdio transport) exposing six x402_* diagnostic tools. Nothing here charges, executes, retries, or repays.

git clone https://github.com/Payloadhq/callx402# no npm install needed for the MCP server

Add this block to your MCP client config (replace the path), then restart the client:

{  "mcpServers": {    "callx402": {      "command": "node",      "args": ["/absolute/path/to/callx402/mcp/index.js"]    }  }}

Ask the client to list its MCP tools: the six x402_* tools should appear. Per-client notes and the raw stdio test handshake: integrations/mcp-clients/.

Note: x402_status works with zero setup. The other five tools dispatch into the v2.0.0 subsystem tree, so they report subsystem_unreachable until CALLX402_V2_ROOT points at the tree (see "Full subsystem commands" above) and the matching flag is enabled.

The relationship

• PAYLOAD = the parent company.
• VEYLINE = production infrastructure for x402 + MCP. The flagship brand.
• CALLX402 BY PAYLOAD = the x402 response/action layer, powered by Veyline. The entry action into Veyline's production infrastructure, not the flagship name. Nothing here renames Veyline.

Discovery path: a developer searching for x402 finds callx402, uses it, and learns Veyline. Payload is not affiliated with the x402 Foundation.

What it does

• Diagnose an x402 or MCP failure
• Rescue a failed transaction (auth-gated)
• Route a paid agent job to the cheapest viable path
• Resolve settlement state from on-chain evidence
• Execute under an explicit budget, with fail-closed safety rules

Behavioral usage language (not trademark claims):

• "Need to diagnose x402? callx402."
• "Need to rescue a transaction? callx402."
• "Need to route a paid agent job? callx402."
• "Need settlement certainty? callx402."
• "Need to execute under a budget? callx402."

Quickstart

1. Install

npm install -g callx402

Or locally in a project: npm install callx402 (the binary is then at node_modules/.bin/callx402).

Fallback, install direct from GitHub:

npm install Payloadhq/callx402

Local development:

git clone https://github.com/Payloadhq/callx402cd callx402npm installnpm link        # exposes the `callx402` command

2. First run: what works with zero setup

callx402 statuscallx402 'complete this job for under $1'

status is the honest starting point: a per-subsystem reachability report. On a bare npm install it reads 0/15 subsystems reachable, which is expected (see step 3). The intent line plans in plain language and stops before execution, so it is always safe to run; use single quotes so your shell does not eat $1.

Free read-only checks against the live Payload rail (no account, no keys):

curl https://payload-rail.fly.dev/v1/callx402/actionscurl 'https://payload-rail.fly.dev/v1/callx402/quote?action=diagnose'

These return the paid action catalog with prices and a free price quote for one action. Invoking a rail action is paid per action via checkout (machine front door: https://payloadhq.github.io/agents.json); nothing here pays anything.

3. Full subsystem commands

diagnose, rescue, route, resolve, doctor, monitor, preflight, and inspect dispatch into the v2.0.0 subsystem tree (the Veyline Developer Primer, x402-paid-api-starter-kit), a separate checkout that is not an npm dependency. Without it these commands exit 3 with subsystem_unreachable and do nothing. If you have the tree, point at it:

export CALLX402_V2_ROOT=/path/to/x402-paid-api-starter-kit/v2.0.0callx402 status    # now: 15/15 reachable, 0/15 enabled

(The default is a x402-paid-api-starter-kit/v2.0.0 directory sitting next to your callx402 checkout.)

Subsystems are disabled by default; status names the flag that enables each one. Set a flag, then run its command:

export PAYLOAD_MCP_DOCTOR=1callx402 diagnose --target 'https://api.example.com/x402/pay'

More examples (each needs its subsystem flag enabled; run callx402 status to see the flag names):

callx402 rescue --incident inc_123callx402 route --goal 'fetch 100 product prices for under 0.50 USD'callx402 resolve --evidence '{"txHash":"0xabc..."}'

Note the single quotes: the goal text contains $-style amounts that a shell would expand inside double quotes.

Alternative to the local tree: run in remote mode against your own callx402 server (server/index.js):

export CALLX402_MODE=remoteexport CALLX402_REMOTE_URL=http://127.0.0.1:8787

JavaScript SDK

const { callx402 } = require('callx402');
const result = await callx402({  intent: 'complete this job for under $1',  maxBudget: 1.00,  networks: ['base'],  assets: ['USDC'],  approvalThreshold: 5.00,  idempotencyKey: 'job-42',});

Python SDK

import callx402
result = callx402.callx402(    intent="complete this job for under $1",    max_budget=1.00,    networks=["base"],    assets=["USDC"],    approval_threshold=5.00,    idempotency_key="job-42",)

HTTP server

CALLX402_PORT=8787 node server/index.jscurl -X POST http://127.0.0.1:8787/call \  -H 'Content-Type: application/json' \  -d '{"intent":"complete this job for under $1","maxBudget":1.00,"dryRun":true}'

Full HTTP surface: server/openapi.yaml (served live at GET /openapi.json).

Commands

Subsystem commands (diagnose, rescue, route, resolve, doctor, monitor, preflight, inspect) need the v2.0.0 tree plus the matching feature flag (see "Full subsystem commands" above); without them they exit 3 and do nothing. status, intent mode, and config work with zero setup.

Paid one-off diagnostics (no subscription)

The CLI and MCP tools above are the free local tier: read-only diagnostics that never charge, never execute, and never move money. When a free diagnostic is not enough — you need the production rail to resolve a settlement, judge a specific retry plan, or triage an incident end to end — each action is also available as a paid one-off rail action. No subscription, no account: quote, then pay deliberately.

The quote-before-payment flow, verified live:

# 1. Free quote: exact price before anything is paidcurl "https://payload-rail.fly.dev/v1/callx402/quote?action=resolve&path=x402"# -> {"quote_id":"...","quoted_price_usd":"0.25", ...}
# 2. Invoke unauthenticated -> x402 v2 402 with the exact payment termscurl -X POST https://payload-rail.fly.dev/v1/callx402/actions/resolve \  -H 'Content-Type: application/json' -d '{"evidence":"{...}"}'# -> HTTP 402: exact USDC amount on Base, pay-to address, 300s window, quote_id
# 3. Pay deliberately from an authorized wallet, then retry with {txHash, quote_id}

Live fee schedule: GET https://payload-rail.fly.dev/v1/callx402/actions (responds "model":"paid on-demand per action; no subscription required"). The x402-path price is the Stripe-path fee divided by 20 — for example resolve is $5.00 via Stripe, $0.25 via the x402 path. Never blind-retry a payment to reach a paid action: get the quote first.

Problem to action map (which paid action answers which failure, with the free CLI/MCP path for each): docs/problem-map.md.

Integrations

Working entry points for agent frameworks, automation, MCP clients, and x402 facilitators — all in integrations/ and tested against the live rail:

• LangChain — 9 tools (integrations/langchain/): diagnose, recover, resolve, evidence, explain, safe-retry, duplicate-payment risk, preflight, plus the free live fee schedule
• CrewAI — the same actions as CrewAI Tools (integrations/crewai/)
• n8n — importable incident-guard workflow (integrations/n8n/): maps an incident to a callx402 action, fetches the free live quote, invokes when credentialed, otherwise emits payment instructions — never auto-pays
• MCP clients — Claude Desktop / Cursor / Windsurf setup for the free read-only MCP server (integrations/mcp-clients/)
• x402 facilitators — settle-guard.js (integrations/facilitator/): resolve settlement state from evidence before re-broadcasting a payment

Problem → action map: docs/problem-map.md. Machine front door for agents: https://payloadhq.github.io/agents.json.

Money-safety rules

• Settlement UNKNOWN is never auto-retried and never repaid.
• Over-budget intents are refused with zero side effects.
• Approval thresholds fail closed.
• Idempotency keys dedupe — repeats return the original, never re-execute.
• Disabled or unreachable subsystems fail closed — success is never faked.
• Non-custodial — callx402 dispatches work; it never holds funds or private keys.

What callx402 is not

• Not the flagship product name. The product is Veyline.
• Not a broker, negotiator, or custodian. It dispatches; it never holds funds.
• The phrases above are usage language, not exclusivity claims.

Links

• Canonical docs: https://payloadhq.github.io/
• Payload org profile: https://github.com/Payloadhq/Payloadhq
• RevRule by Payload (separate product): https://github.com/Payloadhq/revrule-console

Full subsystem reference: docs/README.md. Behavioral language: docs/ACTION_LANGUAGE.md. Design spec: SPEC.md.

License

MIT. See LICENSE.

More from Payload · payloadhq.github.io · all Payload repos

Related: x402-manifest-check · x402-observatory · flow-agentic-demo