MCP-compatible agents
3 progressive tools discover and invoke 24 modules.
Connection guide →
Add one free, no-auth reliability sidecar to an existing agent. Keep the agent you already use; call these tools only when planning, coordination, recovery, or evidence will help.
Remote MCP
These links carry a privacy-safe source label so aggregate initialization and invocation health can be measured. Prompts, tool arguments, credentials, and wallet details are not collected for attribution.
3 progressive tools discover and invoke 24 modules.
Connection guide →
37 direct tools across the same 24 modules.
Connection guide →
37 direct tools across the same 24 modules.
Connection guide →
Access
Every public module is free today. Optional higher-volume x402 execution is staged but inactive.
All public modules appear as direct, no-auth tools at /mcp/claude and /mcp/chatgpt. Each exposes 37 direct tools across 24 modules and has a separate 60-call-per-client, 600-call-global minute budget plus lower per-module limits.
Discovery, schemas, status, validation, unsupported lookups, capability requests, and bounded connector access remain free. Connectors do not collect or transfer payment.
Documentation →A later vendor-neutral HTTP request may buy capacity under explicit Base USDC terms. It will not hide better answers. Prices, limits, receiver, replay, and refund terms must be published before activation.
Not active| Surface | Public boundary | Verification | Distribution status |
|---|---|---|---|
| MCP-compatible agents | 3 progressive tools discover and invoke 24 modules | Automated protocol, contract, discovery, and hosted boundary checks | Published in the Official MCP Registry |
| ChatGPT | 37 direct tools across the same 24 modules | Automated contract and hosted boundary checks; human review materials prepared | Public endpoint; marketplace review is not claimed |
| Claude | 37 direct tools across the same 24 modules | Automated contract and hosted boundary checks; host review remains separate | Public endpoint; directory approval is not claimed |
| Codex and skills-compatible agents | Repo-scoped skill; no persistent MCP; direct HTTP only after local risk selection | 200-run preregistered Codex publication passed; safety gains appeared in overlapping workers and stale scheduled refreshes | Skills-only archive and source are published on GitHub |
Contract-tested means the published protocol boundary, schemas, and hosted behavior passed the project's automated checks. It does not imply approval by a host marketplace or directory.
Open source
Start with this common JSON shape and adapt the surrounding configuration filename to your MCP host.
Inspect the versioned manifest →
List the seven workflow skills before choosing which ones to install.
Inspect all skills →
Reproduce the general reliability-sidecar and multi-agent checkpoint workflows, then adapt them to your framework.
Public claims are tied to reproducible checks. Review the live effectiveness report, measurement methodology, and security policy. New framework and marketplace names are added to this page only after their integration is reproducible.