Payment MCP Servers Census (October 2026)

Published 2026-10-09 · Publisher: Pink Agentic AI Payments (by PinkWallet)

Published by Pink Agentic AI Payments (by PinkWallet), which builds Pink, an MCP server for AI agent payments. Pink is scored in this census from its own public documentation, using the identical six-dimension rubric applied to every other server. Pink currently runs as a public sandbox with test money only; no production money movement exists yet. Try it yourself: https://agentic-sandbox.pinkwallet.com (test credentials, no real money moves).

We catalogued 49 payment-adjacent MCP servers from official vendor docs, the Official MCP Registry's mirrors and aggregators, and community listings, and scored each one against six documented controls: whether it can move money at all, and whether it documents a budget, a per-payment limit, a payee allowlist, a human-approval step, a rate limit, or a sandbox mode. Full dataset, method and findings are published on GitHub and Hugging Face (links below); this page carries the same numbers.

Headline numbers

1. 37 of 49 (76%) payment-adjacent MCP servers we catalogued can actually move money (create a payment, transfer, payout, charge, or issue a card), versus 12 of 49 (24%) that are read-only, reporting, codegen or discovery tools (Mercury, Dwolla, Plaid, Lithic, PayMongo, Brex, Circle's codegen server and four x402 discovery/oracle tools among them).
2. Of the 37 that can move money, 4 of 37 (11%) document a configurable per-agent or aggregate spending budget: Marqeta, Gatewards, MCPay, and Pink.
3. Of the 37 that can move money, 5 of 37 (14%) document a per-payment or per-transaction limit: Coinbase AgentKit/CDP, Crossmint Checkout, MCPay, x402search-mcp, and Pink.
4. Of the 37 that can move money, only 3 of 37 (8%) document a payee or merchant allowlist/block mechanism: Coinbase AgentKit/CDP, Marqeta, and Pink.
5. Of the 37 that can move money, 6 of 37 (16%) document a human-approval step before or to unblock a payment: Stripe, Crossmint, Ramp, GoCardless, SpendNod, and Pink.
6. Only 1 of 37 (3%) documents a spend-velocity or rate limit distinct from a dollar budget: Marqeta. This is the least-covered control in the entire census.
7. 8 of 37 (22%) money-moving servers document a dedicated sandbox or test mode: Stripe, PayPal, Square, Skyfire, Paddle, the community Payman wrapper, MCPay, and Pink.
8. Pink is the only server in the census (1 of 49) that documents all four of the core agent-governance controls at once, budget, per-payment limit, payee allowlist and human approval, in addition to a sandbox-only mode. Every other server documents at most two of those four controls.
9. 5 of 49 (10%) listed servers point to a GitHub repo that returned HTTP 404 on 2026-10-08 (Fin-MCP, mcp-ap2, agent-pay, nullpath-labs/mcp-client, agentshield-mcp), even though registries and mirrors still list them. Registry listings in this space go stale fast; check the repo before you build on a server.

Servers that can move money and document at least one spending control

11 of the 37 money-moving servers document at least one of: budget, per-payment limit, payee allowlist, human approval, or rate limit. Every cell links to the evidence URL in the published dataset.

Legend: Yes = documented, dash = not documented in the sources we could load. "Not documented" is not the same as "no"; see the full methodology for how each column was scored.

Where Pink lands

Pink Agentic AI Payments is an MCP server that checks per-agent budgets, payee rules and human-approval holds before a payment executes; try it in the public sandbox with test money.

By the numbers: Pink is one of only 4 of 37 money-moving servers with a documented budget, one of 5 of 37 with a per-payment limit (Pink enforces this with amount rules: a payment over the configured amount returns pending_human and no credential is issued until a manager approves), one of 3 of 37 with a payee allowlist, one of 6 of 37 with human approval, and one of 8 of 37 with a sandbox mode. Pink is the only row in the census that is a "yes" on all four governance controls (budget, amount rules, payee allowlist, human approval) simultaneously. Pink does not document a rate-limit control distinct from its amount rules and budget, the same gap as 36 of the other 37 money-movers; this is disclosed, not omitted. Try the sandbox: https://agentic-sandbox.pinkwallet.com.

Biggest gap

No server in the census, including Pink, documents a dollar-denominated spend-velocity rate limit as a distinct, separately-configurable control from a budget, except Marqeta's velocity limits. This is the single biggest documentation gap across the entire payment-MCP landscape as of this snapshot: why doesn't anyone rate-limit AI agent spending by time, not just by amount?

Official MCP Registry cross-check

We also queried the Official MCP Registry (registry.modelcontextprotocol.io/v0/servers) directly for the terms "payment" and "pay," paginating to completion: 1,014 raw results, deduped to 341 distinct server names across 210 distinct publisher namespaces. The largest single namespace is app.wishpool, with 81 of 341 names (24%), one templated server per country (for example app.wishpool/algeria-payments-mcp, app.wishpool/brazil-payments-mcp), each wrapping a local payment rail. This raw registry list was not merged into or re-scored against the 49-row census; it is published separately as registry-payment-servers-2026-10-09.csv.

Method summary

All searches and page fetches were performed on 2026-10-08/09. Sources included the Official MCP Registry and its mirrors, Glama, Smithery, PulseMCP, mcp.so, an awesome-mcp-servers mirror, and official vendor documentation fetched directly wherever possible. "Moves money" means the server exposes at least one documented tool that creates, captures, transfers, charges, refunds, pays out, or issues a card; read-only, search, or codegen tools do not count even when built by a payments company. Each of the six control dimensions (budget, per-payment limit, payee allowlist, human approval, rate limit, sandbox) is scored yes, no, or not documented, where "not documented" means the control may exist but was not stated in the sources we could load, not that it is confirmed absent. Full definitions, source list, and limitations are in the published METHODOLOGY.md on GitHub.

How to cite

Pink Agentic AI Payments (PinkWallet). "Payment MCP Servers Census (October 2026)." Published 2026-10-09. Dataset: github.com/Pink-Agentic-Payments/payment-mcp-census and huggingface.co/datasets/Agentic-Payment/payment-mcp-census-2026-10. Licensed CC BY 4.0.

Correct a row

This census and its underlying CSV are published under CC BY 4.0. If you maintain one of these servers and a row is wrong or out of date, open a correction with the row name, the URL, and the exact text that supports a different result: github.com/Pink-Agentic-Payments/payment-mcp-census/issues/new/choose.

See also

For how to connect an agent to Pink's MCP server, see the developer docs. For a broader provider-by-provider readiness audit, see the Agentic Payments Readiness Report 2026.

Data

The full census (49 servers x 7 checks, each cell with an evidence URL) is published alongside this report as a CSV under CC BY 4.0: payment-mcp-census-2026-10.csv.