English Français العربية

io.github.MNKIAgentOS/mnki-mcp

Official registry AI and agents

MCP proxy that checks every tools/call against Agent Trust policy, with evidence, before it runs.

Source code Website Documentation

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

NPM package mnki-mcp
Claude Code 
claude mcp add mnki-mcp -e 'MNKI_API_KEY=<your mnki_api_key>' -e 'MNKI_URL=https://mnki.com' -- npx -y mnki-mcp@0.3.0 --agent '<invoice-agent>' --server '<github>' --mode --upstream '<-- npx -y @modelcontextprotocol/server-github>' --upstream-url '<https://mcp.example.com/mcp>'
JSON configuration (Claude Desktop, Cursor, VS Code) 
{
    "mcpServers": {
        "mnki-mcp": {
            "command": "npx",
            "args": [
                "-y",
                "mnki-mcp@0.3.0",
                "--agent",
                "<invoice-agent>",
                "--server",
                "<github>",
                "--mode",
                "--upstream",
                "<-- npx -y @modelcontextprotocol/server-github>",
                "--upstream-url",
                "<https://mcp.example.com/mcp>"
            ],
            "env": {
                "MNKI_API_KEY": "<your mnki_api_key>",
                "MNKI_URL": "https://mnki.com"
            }
        }
    }
}

Environment variables

  • MNKI_API_KEY required secret Agent Trust API key with the verify scope (console → Settings → API keys).
  • MNKI_URL Console base URL; https://mnki.com unless self-hosted.

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Agent Trust

Prove what an autonomous AI agent is authorized to do. An open profile plus a reference implementation for agent identity, principal binding, delegation with attenuation, capability-based authorization, request proof-of-possession, authorization attestations, tamper-evident provenance, single-use permits, cross-organization revocation and declared effect paths — carried on JWS, OAuth/OIDC, SPIFFE/WIMSE, AuthZEN, OpenID Federation, MCP and A2A. No new cryptography, no new protocol.

No consequential agent action should be accepted because a process is running or a token is valid. It should be accepted because the action is attributable, authorized, constrained, current and verifiable.

PackageWhat it is
packages/verifierThe pure core: parsing, delegation chain + attenuation, policy engine, request proof, delegation credentials, attestations, the 12-step evidence pipeline. Deterministic, dependency-free, WebCrypto only.
packages/sdk-tsmnki-sdk — TypeScript SDK: agent identity (keys, signed requests), guard() for any tool call, verify, delegate, attest, request grants from an access broker, offline verification against an organization's JWKS; framework adapters as subpaths.
packages/climnki-cli — mnki demo · init · identity · verify · inspect · delegate · attest · access · scan · protect · conformance (agenttrust is an alias).
packages/mcpmnki-mcp — a local MCP proxy that verifies every tools/call before it reaches the server (mnki protect installs it).
python/pip install mnki — Python SDK (zero dependencies; mnki[signing] for Agent-Proof).
go/Go verifier (same vectors, no dependencies) and at-verify, the self-hosted verifier that serves /v1/verify from an exported organization snapshot.
spec/Agent Trust Profile v0.2, the interoperability positioning, the Internet-Draft (draft-mnki-agent-trust-profile-00) and the standards plan.
conformance/Executable vectors for certification levels 1–3 (LEVELS.md) — the cross-language contract; the TypeScript, Go and Python runners pass them all.
npx mnki-cli demo      # the 15-second story, no account: a €47,000 refund denied, €420 allowed
npx mnki-cli scan      # find the MCP servers and API keys agents can use on this machine
npm install mnki-sdk   # guard() any tool call
pip install mnki               # same in Python

Develop against the open verifier and SDKs for free; govern fleets with the hosted control plane at mnki.com — see COMMERCIAL.md for the boundary and TRADEMARK.md for name and badge use. Releases carry npm/PyPI provenance, SBOMs and cosign-signed checksums (SECURITY.md).

Apache-2.0.

Frequently asked questions

What is io.github.MNKIAgentOS/mnki-mcp?

MCP proxy that checks every tools/call against Agent Trust policy, with evidence, before it runs.

How do I install io.github.MNKIAgentOS/mnki-mcp?

It runs on your machine. In Claude Code: claude mcp add mnki-mcp -e 'MNKI_API_KEY=<your mnki_api_key>' -e 'MNKI_URL=https://mnki.com' -- npx -y mnki-mcp@0.3.0 --agent '<invoice-agent>' --server '<github>' --mode --upstream '<-- npx -y @modelcontextprotocol/server-github>' --upstream-url '<https://mcp.example.com/mcp>'. For other clients, paste the JSON configuration from this page into their MCP settings.

Does io.github.MNKIAgentOS/mnki-mcp need an API key?

Yes. It expects MNKI_API_KEY, which you set in its configuration.

Is io.github.MNKIAgentOS/mnki-mcp open source?

Yes. Its code is on GitHub under the Apache-2.0 license.

Who maintains io.github.MNKIAgentOS/mnki-mcp?

It is published by MNKIAgentOS. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.