Source: raw/x-account-nousresearch-2066647737613832624.md

Author: @NousResearch (Nous Research) | URL: https://x.com/NousResearch/status/2066647737613832624 | Posted: 2026-06-15

In partnership with Stripe, Hermes Agent now ships a first-party suite of Stripe payment skills, giving the agent real-world spending power — buying on the open web, paying per-call for APIs, and provisioning its own SaaS. The skills install from the official catalog under the official/payments/ namespace and carry a built-in safety model: configurable human-approval gates, credentials kept out of the transcript, and one-time credentials cleaned up after use. This is the payments counterpart to the vetted-install discipline Hermes already applies to MCP servers and skill bundles.

Evidence note: this is a first-party @NousResearch announcement; the capability and safety claims are vendor assertions, and the raw capture is an x_search synthesis of the post — exact wording, media, and per-skill specifics should be confirmed against the linked post and the docs catalog before relying on details.

Key Takeaways

  • Agent commerce, three capabilities. The Stripe skill suite lets Hermes (1) buy things on the open web, (2) use pay-per-call APIs, and (3) provision its own SaaS — the agent can transact, not just read.
  • First-party, with a Stripe partnership. Shipped by Nous Research “in partnership with Stripe” — these are official skills, not community wrappers.
  • Three named skills. stripe-link-cli, mpp-agent, and stripe-projects, installed individually from the official catalog.
  • Install pattern. hermes skills install official/payments/<skill> — the same hermes skills install surface as the rest of the official skills catalog, scoped to the official/payments/ namespace.
  • Built-in safety model — three controls:
    • Configurable human-approval gates — spend actions can require human sign-off.
    • Credentials never enter the transcript — payment secrets stay out of the model’s context and logs.
    • One-time credentials cleaned up after use — ephemeral credentials are removed post-transaction.
  • Docs home. The payments skills live in the optional-skills catalog: hermes-agent.nousresearch.com/docs/reference/optional-skills-catalog#payments.
  • Consistent with the rest of Hermes. The approval-gate + credential-isolation model extends the same least-privilege discipline documented in the Hermes Security Model, which already names the Stripe key as a credential kept invisible to tool subprocesses.

Try It

  1. Browse the catalog. Read the payments section of the optional-skills catalog: hermes-agent.nousresearch.com/docs/reference/optional-skills-catalog#payments.
  2. Install a payment skill:
    hermes skills install official/payments/stripe-link-cli
    hermes skills install official/payments/mpp-agent
    hermes skills install official/payments/stripe-projects
  3. Set the approval gate before any live spend. Configure the human-approval gate so transactions require sign-off until you trust the workflow.
  4. Verify credential hygiene. Confirm payment credentials never appear in the transcript and that one-time credentials are cleaned up after use before running an unattended commerce loop.
  5. See the install follow-up post: https://x.com/NousResearch/status/2066647740801593377

Open Questions

  • mpp-agent acronym. The source names the skill but does not expand “MPP”; the distinct responsibilities of mpp-agent vs stripe-link-cli vs stripe-projects are not detailed in the post.
  • Release binding. Which Hermes version the payments suite ships in is not stated in the source.
  • Provenance. The raw capture is an x_search synthesis of the @NousResearch post; exact wording, per-skill scope, and the approval-gate configuration mechanics should be confirmed against the linked docs catalog and post before reproduction.