npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@intyga/mcp-proxy

v3.1.1

Published

Stdio proxy that gates a third-party MCP server's tool calls behind a signed human approval — no code changes to the wrapped server.

Readme

@intyga/mcp-proxy — human approval in front of any MCP server, no code changes

Release gated by INTYGA

A stdio proxy that sits between an MCP client (the agent) and a third-party MCP server. Every tools/call is checked against your local policy; a deny is refused locally and every other result goes through your INTYGA gateway's authorization flow. AI-agent execution also requires a v1 receipt and RP check of the human approval. The wrapped server is not modified and is deliberately not trusted.

For an AI_AGENT key, pass --agent-v1-module pointing to an RP-owned JavaScript module whose default export implements AgentV1Runtime from @intyga/mcp-sdk. The proxy then sends the v1 agentContext, verifies the complete receipt against RP-pinned approvers and live configuration, and requires an atomic nonce/session/budget reservation before forwarding the call. Without that module the gateway refuses AI-agent requests.

For a SERVICE or human key, pass --execution-trust-module. Its default export supplies independent tenantId, requesterDid, approvers, requirement, and optional verifier options. Exactly one trust module is required; gateway status alone never authorizes forwarding.

intyga-proxy \
  --target-command npx \
  --target-args '["-y", "@modelcontextprotocol/server-filesystem", "/data"]' \
  --agent-id did:intyga:agent-fs \
  --target fileserver-prod \
  --agent-v1-module ./rp-agent-v1.mjs \
  --local-policy ./policy.json

Point your MCP client's stdio server config at intyga-proxy instead of the real server; the proxy spawns the real server as a child and mediates the protocol stream.

Flags and environment

| Flag | Env var | Meaning | | :--- | :--- | :--- | | --gateway-url | INTYGA_GATEWAY_URL | INTYGA gateway (default http://localhost:8787). Must be https://; plain http:// is refused at startup except to a loopback host (localhost, 127.0.0.0/8, ::1) | | --client-id | INTYGA_CLIENT_ID | Agent credential id | | --client-secret | INTYGA_CLIENT_SECRET | Use the env var. The flag is accepted but warns: argv is world-readable via ps, including by the wrapped server — the one process this proxy exists to distrust | | --agent-id | INTYGA_AGENT_ID | Default value for --target when that is unset. Never sent to the gateway — the identity the ledger records as requester comes from the --client-id credential exchange | | --target | INTYGA_TARGET | Relying-party identifier approvals are bound to (DIV Target Isolation). Defaults to the agent id — never to the gateway's "global" fallback | | --enforcement | — | local-first (default) or gateway-enforced (every call escalates) | | --local-policy | — | Path to a policy manifest JSON — same format and same evaluator as @intyga/mcp-sdk, so the two enforcement paths cannot disagree | | --target-command | — | The real MCP server executable (required) | | --target-args | — | Its arguments, as a JSON array of strings — an object is refused, because Node would read it as the spawn options and a {"shell": true} smuggle would turn the command into a shell string | | --agent-v1-module | INTYGA_AGENT_V1_MODULE | Absolute or relative path to the trusted RP v1 runtime module. Mandatory for AI-agent keys. Export default with tenantId, requesterDid, approvers, requirement, verifier, prepare, liveConfig, readState, and reserve as documented in @intyga/mcp-sdk. | | --execution-trust-module | INTYGA_EXECUTION_TRUST_MODULE | RP-owned SERVICE/human receipt trust. Export default with tenantId, requesterDid, approvers, requirement, and optional verifier. Mutually exclusive with --agent-v1-module. |

Security properties (each of these is tested)

  • Fail closed, everywhere. A line that is not valid JSON is refused with a JSON-RPC -32700, never forwarded — lenient parsers on the far side (Python's json.loads accepts NaN) can read a "non-JSON" line as a tool call this proxy never saw. A throw anywhere in the gate refuses the call; it never falls back to forwarding.
  • JSON-RPC batches are refused (-32600). A batch has no top-level method, so it would classify as non-tool traffic while carrying any number of ungated tools/call entries. MCP removed batching in revision 2025-06-18; one request per line.
  • The approved bytes are the executed bytes. Messages are re-serialized from the parse this proxy evaluated, so a duplicate-key payload cannot read as ping here and as a tool call to a target server whose parser resolves duplicates differently. An approved tools/call is rebuilt from exactly the approved name and arguments (plus _meta.progressToken, which only correlates progress notifications); any other _meta key, params field (including MCP task augmentation) or top-level member is dropped. A tools/call whose name is not a string, or whose arguments is present but not a JSON object, is refused with -32602 before any approval is requested. Omitted arguments becomes {} for both approval and execution.
  • Gated calls run one at a time, FIFO. Two mutating calls cannot be reordered around each other, and a runaway agent cannot fan out unbounded approval prompts. Non-mutating protocol traffic (initialize, ping, */list) bypasses the queue so a pending human decision cannot make the client conclude the server is dead.
  • Every call requires a verified, human-signed receipt. AUTO_APPROVED is refused. Both module types bind tenant, requester, target, parameters, nonce, approver trust and requirement floor before one-time gateway consumption. v1 agent calls also require an RP reservation. Even a local allow rule cannot skip that check, including when --agent-v1-module was omitted. The gateway refuses an AI-agent key without its v1 context. reserve must use a durable compare-and-swap transaction and enforce the budget across sessions. The proxy does not provide a database or a trustworthy view of the agent's actual model, tools or prompt; your RP module must provide both. Executable floors support requiredApprovals, requesterCannotApprove, requireHardwareKey, and excludedSignerDids; fields such as allowedAaguids and signerClass are rejected at startup rather than silently ignored. Use requireHardwareKey for direct model restrictions, subject to the verifier's documented offline assurance limits.
  • The wrapped server never sees your INTYGA credentials. Every INTYGA_* variable is stripped from the child's environment; everything else passes through. Run untrusted third-party servers under a separate OS identity or sandbox so they cannot read the RP module's state, files or parent process. Environment filtering alone is not an OS security boundary.

Runtime

The published package ships compiled JavaScript (dist/cli.js); npx @intyga/mcp-proxy and a global install both run it with no toolchain of their own. Node ≥ 24.

It used to ship the TypeScript entry point directly and rely on Node’s type stripping. That cannot work from an installed package: Node refuses to strip types for any file under node_modules (ERR_UNSUPPORTED_NODE_MODULES_TYPE_STRIPPING), at every version, so the only entry point failed on first run. Working from a clone is unaffected — tsx src/cli.ts … still runs the source.

License

Apache-2.0 — see LICENSE.