Agent Commerce Mcp — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Agent Commerce Mcp (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Agent-native storefront for 14 SaaS/dev/service products and 9 MCP servers. Discovery, pricing, authenticated checkout, affiliate program (15-30% recurring), AgentTrust verification — all over one MCP server.
This is a working A2A commerce layer. Other AI agents can anonymously discover, query, and compare products in the Halverson IQ portfolio. Stateful write tools are only exposed when the server is configured with a write token.
Claude Desktop / Cursor / Cline — add to your MCP config:
{
"mcpServers": {
"agent-commerce-mcp": {
"command": "npx",
"args": ["-y", "agent-commerce-mcp"]
}
}
}Claude Code:
claude mcp add commerce npx -y agent-commerce-mcpRemote (Streamable HTTP):
https://commerce.halversonco.com/mcp(Custom domain commerce.halversonco.com resolves once DNS is wired.)
Public read-only tools are anonymous by default. Set AGENT_COMMERCE_WRITE_TOKEN (or legacy MCP_WRITE_TOKEN) to advertise and execute stateful tools:
create_checkoutregister_affiliaterequest_partnershiprequest_product_consultationWhen the write token is absent, those tools are not advertised and direct calls are rejected fail-closed. When it is present, HTTP clients must send either:
Authorization: Bearer <AGENT_COMMERCE_WRITE_TOKEN>or:
X-Agent-Commerce-API-Key: <AGENT_COMMERCE_WRITE_TOKEN>Set SALESBOT_LEAD_CAPTURE_URL to the Salesbot POST /leads/capture URL to enable opt-in consultation capture. If it is unset, consultation requests fail closed and are not reported as captured.
The remote HTTP server rate-limits MCP requests per client and bounds active MCP sessions in memory. By default it uses the direct socket IP. Set TRUST_PROXY=true only when exactly one trusted reverse-proxy hop sits in front of the process; Express then derives req.ip using that one-hop policy instead of trusting forwarding headers directly.
Discovery
search_products(query, category?, budget_max?, use_case?) — ranked matchesget_recommendation(problem, stack?, company_size?) — 1-3 best-fit picks with reasoningcompare_products(slugs, vs_competitor?) — feature/price matrixPurchase
get_pricing(product_slug, tier?, billing?) — full breakdowncreate_checkout(product_slug, tier, email, referral_code?) — authenticated checkout URLget_free_tier(product_slug) — instant access (signup URL or install command)get_mcp_install(product_slug, client) — exact install snippet for claude_desktop / claude_code / cursor / cline / windsurfBuyer consultation (explicit opt-in only)
request_product_consultation(product_slug, email, name?) — request contact about AgentFetch, QueryShield, or Agent Commerce; call only after the buyer explicitly asks to be contactedAffiliate
get_affiliate_info(product_slug?) — commission rates and tiersregister_affiliate(agent_id, operator_email, products?) — instant signup, returns referral coderequest_partnership(proposal, agent_id, contact_email, integration_type?) — partnership pipelineCross-sell & Trust
get_cross_sells(current_product, agent_context?) — graph-driven adjacencies + recommended bundleget_trust_score() — wraps AgentTrust MCP for vendor reputationverify_vendor() — vendor info, products live, refund/data policiescommerce://catalog/all
commerce://catalog/saas
commerce://catalog/developer
commerce://catalog/services
commerce://catalog/mcp-servers
commerce://product/{slug}
commerce://bundle/{slug}
commerce://affiliate/programelevator_pitch · full_pitch · objection_handler · bundle_pitch
SaaS: CastIQ · GrantIQ · FocusIQ · Catholic Daily Developer: AgentFetch · QueryShield · InjectShield · ModelWatch · ComplianceBeacon · RegImpact Services: Branded Audits · LeadVault · JobAuditor · Halverson IQ Digital Library MCP Servers: AgentFetch · GrantIQ · OutdoorIQ · PubRecords · QueryShield · InjectShield · ModelWatch · Agent Commerce · AgentVault
GrantIQ and PubRecords are part of Halverson IQ Data; their existing URLs stay in place because no separate listing URL is published yet. OutdoorIQ is the CastIQ API tier. ToolPulse tool monitoring is part of ModelWatch. AgentVault folds into the QueryShield agent security suite. The AI Security Stack bundle still points at QueryShield and InjectShield.
Bundles: AI Security Stack (20% off) · Agency Growth Kit (15% off) · AI Builder Essentials (15% off)
15-30% recurring commission on every successful checkout. Tier up by referral count. AI agents register with one tool call and get a referral code immediately.
agent-commerce/
├── server/ # TypeScript MCP server (stdio + Streamable HTTP)
├── catalog/ # products.json, bundles.json, cross-sell-graph.json, affiliates.json
├── outreach/ # Daily registry submission + partner pipeline (Python)
├── analytics/ # Tool-call funnel + Discord daily report (Python)
├── salesbot-integration/ # Drop-in modules for the Halverson IQ salesbot
├── Dockerfile
├── railway.json
└── README.mdMIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.