Chp

io.github.capabilityhostprotocol由 GitHub 账号 capabilityhostprotocol 发布v0.8.0更新于 Oct 25, 2026

Learn and adopt the Capability Host Protocol — the agentic web's open evidence layer.

已验证Streamable HTTP可网页运行知识与记忆开发者工具安全与监控

概览

AI 生成的概览

Capability Host Protocol 的远程端点,作为开放证据层,记录并回放智能体、工具和系统做了什么以及受什么治理

可以做什么
通过 capabilityhostprotocol.com 上的托管端点了解并采用 Capability Host Protocol;清单未列出该端点自身的工具,因此助手主要获得协议上下文而非可调用操作。协议本身定义能力与主机描述符、调用信封、关联上下文、结构化执行证据、结果、错误与拒绝语义,以及回放查询。证据事件以哈希链方式记录并保存在本地 SQLite 存储中,回放按关联 ID 进行。可对已发生的运行评估策略,显示本应拒绝哪些事件。提供 Python 与 TypeScript 参考实现,以及实验性的 MCP 桥接。
适用场景
适合在你希望了解或采用 CHP,以获得受治理、防篡改的智能体与工具执行记录,或将其与 MCP、OpenTelemetry 比较时使用。它不是通用工具服务器,若只需要可调用工具,价值有限。
运行要求
需要访问 capabilityhostprotocol.com 的远程 streamable HTTP 端点;清单声明无需认证、环境变量或请求头。本地使用该协议需要 Python(pip install chp-core)或 Node.js 的 TypeScript SDK,并将证据写入用户主目录下的本地 SQLite 文件。
安装前请注意
项目为 1.0 之前的候选版本,接口仍可能变化。README 提到强制认证开关(CHP_HOST_REQUIRE_AUTH)和密钥泄露处置手册,说明生产环境中签名密钥与认证配置很重要。托管能力图、保留期、团队工作区与合规导出等能力被描述为商业版而非开源。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Chp,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.capabilityhostprotocol%2Fchp/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "chp": {
      "type": "http",
      "url": "https://capabilityhostprotocol.com/api/mcp"
    }
  }
}

README

Capability Host Protocol

[PyPI] [Python] [npm] [License] [Docs]

CHP is the open protocol for declaring, governing, and proving what agents, tools, and systems do — the single signed plane where a human approval, an agent's action, and a system call become the same governed, tamper-evident, replayable event.

The hook is simple:

See what your agents and tools actually did — and what governed it.

Try it in two minutes

Your coding agent reads files, runs commands, calls tools. This puts a governed boundary at the point of action — no application code changes:

bash
pip install chp-corechp hooks install                        # hooks Claude Codechp hooks install --all-harnesses        # ...or Claude Code + Codex + Gemini CLI

Use your agent normally, then look at what it did:

bash
chp session listchp session tree <session_id>

Every tool call is a typed evidence event, hash-chained and stored locally in ~/.chp/evidence.sqlite. A denial is a first-class event with a reason code — not a swallowed exception — and the chain is tamper-evident, so someone who did not run the agent can still tell whether the record is intact.

Prefer to drive the protocol directly? chp serve-demo starts a governed host and prints a copy-pasteable first invocation.

Full guide: docs/quickstart.md · why it exists: docs/why-chp.md · docs site: docs.capabilityhostprotocol.com

What you get back

Replay is by correlation ID, and the record answers more than "what happened" — from examples/agent-operations-demo/:

json
[  {"sequence": 1,  "event_type": "execution_started",   "capability_id": "trace_execution", "outcome": null},  {"sequence": 3,  "event_type": "execution_completed", "capability_id": "trace_execution", "outcome": "success"},  {"sequence": 7,  "event_type": "execution_started",   "capability_id": "tool.add",        "outcome": null},  {"sequence": 12, "event_type": "execution_started",   "capability_id": "tool.multiply",   "outcome": null},  {"sequence": 13, "event_type": "execution_completed", "capability_id": "tool.multiply",   "outcome": "success"}]

You can also ask what a policy would have done to a run that already happened:

json
{  "invariant": {"id": "deny_multiply_tool", "kind": "capability_id_matches"},  "would_have_denied": true,  "violating_events": [{"capability_id": "tool.multiply", "event_type": "execution_started"}]}

Why a protocol, not a library

CHP is not another agent framework, tool protocol, or workflow engine. It is the governed evidence plane at the capability boundary: what ran and what governed it (policy, risk tier, safety checks, human approval, autonomy budgets, denial) emit onto one signed, correlated record. Observability tools split execution across separate, optional, unsigned signals and carry no governance; CHP unifies both and proves them.

Status: CHP is a pre-1.0 release candidate (v0.9.3) — a frozen, additive wire surface backed by two independent implementations (Python + TypeScript) that pass conformance. chp-core ships on PyPI.

What CHP Defines

  • Capability descriptors
  • Host descriptors
  • Invocation envelopes
  • Correlation context
  • Structured execution evidence
  • Outcome, error, and denial semantics
  • Replay queries and results
  • Replay by correlation ID
  • Minimal conformance requirements

Install

bash
pip install chp-core                 # zero runtime dependenciespip install 'chp-core[schema]'       # enforce declared input_schemapip install 'chp-core[signing]'      # ed25519 — signed hosts, bundles, mandatesnpm install @capabilityhostprotocol/sdk   # TypeScript client + verifier (alpha)

chp host verify smoke-tests the install in under a second and reports whether input-schema validation is enforced.

From this checkout: python -m pip install -e packages/python.

Minimal Capability

python
from chp_core import LocalCapabilityHost, capability
host = LocalCapabilityHost("example-host")
@capability(    id="math.add",    version="1.0.0",    description="Add two numbers.",)def add(a: int, b: int):    return {"sum": a + b}
host.register(add)
result = host.invoke(    "math.add",    {"a": 2, "b": 3},    correlation_id="demo-correlation",)
print(result.outcome)       # "success"print(result.data)          # {"sum": 5}
for event in host.replay("demo-correlation"):    print(event.event_type)  # execution_started, execution_completed

The host emits execution_started and execution_completed evidence for the invocation. If execution fails, it emits execution_failed. If the host denies invocation, it emits execution_denied.

Repository Map

  • spec/README.md: the specification index — core (v0.1), governance vocabulary, invocation pipeline, HTTP binding, evidence integrity + anchors (v0.2), reserved names, test vectors, changelog, proposal process
  • schemas/: JSON Schemas for protocol objects
  • packages/python/chp_core/: reference host (Python)
  • packages/chp-sdk/ + packages/chp-host-ts/: the second implementation (TypeScript)
  • examples/capability-host-endpoint-demo/: HTTP-served host demo
  • examples/agent-operations-demo/: agent/tool observability demo
  • examples/codex-self-observation-demo/: Codex dogfooding demo
  • examples/mcp-bridge-demo/: experimental MCP-style bridge prototype
  • conformance/: conformance runner
  • docs/quickstart.md: install, first run, serving, mesh
  • docs/why-chp.md: the problem and the thesis
  • docs/adapter-authoring.md: writing your own capability adapter
  • docs/production-runbook.md: operations, backup/restore, key compromise
  • docs/comparisons/chp-vs-mcp.md: precise MCP comparison
  • docs/comparisons/chp-and-opentelemetry.md: OpenTelemetry alignment note
  • docs/comparisons/landscape.md: adjacent framework comparison
  • docs/security/threat-model-v0.1.md: v0.1 threat model

Production Posture

The reference implementation is hardened for production operation: WAL multi-writer safety with hot backup (chp store backup --verify), SIGTERM drain (in-flight work completes before exit), a fail-loud auth flag (CHP_HOST_REQUIRE_AUTH=1), scheduled retention, and operator metrics (store size, witness-loop liveness, revocation counts, internal errors). Operations, backup/restore, rolling upgrades, and the key-compromise runbook: docs/production-runbook.md. Vulnerability reporting: SECURITY.md.

CHP vs MCP

MCP exposes tools and context to AI applications. CHP governs and evidences execution of capabilities.

They fit together. MCP can be a source of capability invocation, and CHP adds correlation, replay, evidence, denial semantics, and governance at the execution boundary.

Read more: docs/comparisons/chp-vs-mcp.md.

Contributing

Read CONTRIBUTING.md first. One thing to know before you open a pull request: this repository is a generated mirror of a private development repository. CI rejects pull requests that touch packages/, docs/, spec/, schemas/, or examples/ from a branch that is not sync/*, so a code PR against those paths will fail by construction no matter how good it is.

That is not a brush-off — it is the publishing model, and we would rather say so up front than let you find out from a red check. Issues, spec proposals, comparisons, and discussion are the highest-bandwidth way in today; open an issue and we will route the change through the internal flow with attribution.

Open Source Boundary

Open source should include local visibility:

  • spec and schemas
  • local host
  • SDK primitives
  • conformance
  • local replay
  • agent observability wrapper
  • experimental MCP bridge prototype

Commercial value can remain around production trust:

  • hosted capability graph
  • multi-host trace stitching
  • retention
  • team workspaces
  • advanced explanation
  • invariant libraries
  • assurance derivation
  • compliance exports
  • enterprise identity and RBAC

Guiding rule:

Local visibility should be free. Production trust should be paid.

License

CHP is dual-licensed by asset:

  • Code (packages/, conformance/, examples/, scripts/): Apache License 2.0 — see LICENSE.
  • Specification, schemas & docs (spec/, schemas/, docs/): Creative Commons Attribution 4.0 (CC BY 4.0) — see LICENSE-DOCS. Implementing the specification is additionally covered by a royalty-free patent grant — see PATENTS.
  • Trademarks: "CHP" and "CHP-Certified" — see TRADEMARK.md.

Contributions are accepted under the Contributor License Agreement; see CONTRIBUTING.md.

Copyright © 2026 Project Auxo, Inc. See NOTICE.

来源:README.md,提交 6ffb84a

工具

0
SourceWeft 尚未读取此服务器的工具。

版本历史

1
  1. v0.8.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式必需10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分