rnv-mcp-identity
A deterministic identity and authorization layer for MCP servers that verifies agent identity and scope using OAuth 2.0, WIMSE workload identity, and HTTP Message Signatures, resolving or refusing without guessing.
README
rnv-mcp-identity
An identity and authorization layer for MCP servers. On every tool call it asks one question, is this caller who it claims to be, and is this action within what it's allowed to do?, and answers with exactly one of three outcomes:
- allow: identity resolved, verified, and the action is in scope.
- deny: verified, but out of scope; or an identity was presented and failed verification.
- unknown: identity could not be established at all. The call is refused.
An unknown caller never acts. That rule is the whole project: resolve or refuse, never guess.
Status: reference implementation, pre-1.0, Apache-2.0. This is the reference-implementation arm of AIII, the Artificial Intelligence Identification Initiative. It composes on existing standards (the MCP authorization model, WIMSE workload identity, RFC 7800 key confirmation, RFC 7638 thumbprints, EAT attestation) rather than replacing them.
Live demo: Identity Gate on Hugging Face Spaces, running this engine on the released v0.1.0.
What's here
src/rnv_mcp_identity/— the library: a framework-agnostic decision engine (L1 identity, L2 verification, L3 authorization) plus a FastMCP middleware adapter. The core has no runtime dependencies.examples/— a runnable FastMCP server guarded by the layer, an in-process test, and an HTTP client. Start here to watch it allow one call and refuse three.SPEC.md— the normative spec: the decision sequence, the capability and policy model, the threat model, and the v0 wire format.AAIF-READINESS.md— an honest dossier on whether this belongs in a foundation, and what's missing before it would.
Quickstart
pip install -e ".[dev,verify,fastmcp]"
python -m pytest -q # the full suite
python -m pytest -q tests/test_demo_inprocess.py # just the guarded-server demo
Run the demo over real HTTP:
python examples/demo_server.py # terminal 1
python examples/demo_client_http.py # terminal 2
The wire format (v0)
The identity token rides in the mcp-agent-identity header and a holder-of-key
proof in mcp-agent-proof. The proof binds to the exact token, so a stolen token
alone can't act. The reference client in examples/ is the normative example; see
SPEC.md section 10.
Docs
SPEC.md— the specificationROADMAP.md— where this is goingPRIOR-ART.md— what it builds on, and the gap it fillsGOVERNANCE.md— how decisions are madeCONTRIBUTING.md— how to helpSECURITY.md— how to report vulnerabilitiesAAIF-READINESS.md— foundation-readiness assessment
License
Apache-2.0. See LICENSE.
A note on affiliation
This project aspires to become a foundation-hosted project and is not affiliated with, endorsed by, or accepted into the AAIF or the Linux Foundation. See AAIF-READINESS.md for an honest account of where it stands.
Verifying a release
Each GitHub release carries the build artifacts, a SHA256SUMS manifest, a
Sigstore signature over that manifest (SHA256SUMS.sigstore.json), and a
CycloneDX SBOM (sbom.cdx.json).
Check integrity (hashes):
sha256sum -c SHA256SUMS
Verify authenticity and the signer's identity (keyless Sigstore):
pip install sigstore
python -m sigstore verify identity \
--cert-identity "https://github.com/RNVizion/rnv-mcp-identity/.github/workflows/release.yml@refs/tags/v0.1.0" \
--cert-oidc-issuer "https://token.actions.githubusercontent.com" \
--bundle SHA256SUMS.sigstore.json \
SHA256SUMS
A valid result proves the manifest was produced by this repository's release
workflow under GitHub's OIDC, not by a third party. Replace the tag in
--cert-identity with the release you are verifying.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。