Skip to main content
Glama
zhouwen-giser

SDAR MCP Tasks Provider Runtime

SDAR MCP Tasks Provider Runtime

An independently deployable, language-neutral Runtime for the SEP-2663 task lifecycle and the io.sdar/taskExecution Provider Profile. The Runtime is implemented in strict TypeScript and delegates resource facts and side effects to versioned gRPC/Protobuf Adapters.

The current development target is v2.0.0-rc.1, migrating the primary /mcp endpoint to the frozen SDAR MCP Tasks Unified Protocol Profile V1.0 while retaining the 1.1.0 Provider Ops Wire Schema. Its live plan is frozen-protocol-v1-exec-plan.md. Published v1.0.0-rc.2 and v1.0.0-rc.3 migrations, reports, tags, and release history remain immutable.

中文接口文档导航

面向外部系统的接口以中文说明为主,并保留协议中的英文标识符:

  • Runtime API 参考:说明 HTTP 健康/管理接口、MCP JSON-RPC 方法、tasks/observations(任务观测分页)和 Adapter gRPC 方法,包含输入输出样例。

  • Provider 遥测入口:说明 ProviderTelemetryIngress(Provider 遥测入口)及 EmitProviderEvents(批量提交事件)的字段、mTLS 身份规则、错误码和 TypeScript/Python 客户端样例。

  • Provider 运维遥测:说明 ProviderOpsEnvelope(Provider 运维事件信封)、外部 Collector 输出、指标、隐私和失败语义。

  • Runtime 配置参考:逐项解释环境变量、默认值和生产环境失败关闭要求。

  • Adapter 快速开始:说明 Provider/Adapter 对接方向和双语言示例客户端。

Related MCP server: WorkspaceGuard

Runtime quick start

Prerequisites: Node.js 22, Corepack/pnpm 11, and Docker with Compose access.

corepack enable
pnpm install --frozen-lockfile
pnpm build
pnpm test:unit
pnpm test:contract
docker compose up --build --wait
curl --fail http://127.0.0.1:8080/health/ready

The default stack exposes Runtime HTTP on :8080, PostgreSQL on :5432, and an internal TypeScript Adapter on :7001. A Python Adapter image can be built with:

docker compose --profile python-adapter build adapter-python

Runtime startup applies migrations and runs durable scheduling/recovery before readiness. The reference Adapter state and PostgreSQL data use named volumes. For a release gate with PostgreSQL and Docker available, run:

TEST_DATABASE_URL=postgresql://sdar:sdar@127.0.0.1:5432/sdar_runtime_test pnpm verify:v2

Configuration and security are documented in configuration.md and security-recovery.md; deployment and incident procedures are in the runbook. OTLP signal contracts, privacy rules and failure behavior are in provider-ops-telemetry.md. The signal architecture and Provider-facing gRPC contract are documented in observability.md and provider-telemetry-ingress.md.

Adapter authors should begin with the quick start and dual-language expanded Adapter protocol workflow in adapter-testkit.md. API/RPC and state semantics are summarized in api-reference.md and state-reason-mapping.md. The machine reports mark Runtime Profile coverage partial and real-resource safety not_claimed; a Mock Adapter result is not production qualification. Both mock Adapters include Provider telemetry clients and examples for resource state/metric/health, Task-bound execution progress, and replaying the same Provider event id after an uncertain call. This Provider-to-Runtime service requires no Provider-side OpenTelemetry SDK.

The standalone Home Assistant Climate Provider adds allowlisted climate.* state, power, HVAC-mode, and target-temperature operations through Home Assistant REST and WebSocket APIs.

Production Kubernetes JSON manifests are under deploy/kubernetes, with migration/upgrade instructions in docs/database/upgrade.md. Root commands in package.json expose every release gate; pnpm verify:v2 includes the frozen contract/hash checks, the numbered 74-case report, and the existing formatting, lint, types, build/Proto drift, audit/SBOM, deployment/container, unit/contract/integration/recovery/security/E2E/conformance, the six rc.1 red-regression guards, and the rc.3 capacity checks. CI additionally runs Buf lint/breaking against the immutable rc.1 tag and builds the Runtime plus both Adapter images with Compose.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    A
    maintenance
    Deterministic execution engine for AI agents. 412 modules across 78 categories including browser automation, file I/O, Docker, data parsing, crypto, and scheduling. Supports STDIO and Streamable HTTP transport with execution trace, evidence snapshots, and replay from any step.
    281 PyPI
    481
    Apache 2.0
  • F
    license
    B
    quality
    D
    maintenance
    Structured workspace runtime for long-running coding agents, providing controlled workspace capabilities with task state, snapshots, checkpoints, drift detection, verification evidence, audit logs, and structured handoff.
    20
    -
  • F
    license
    Not graded
    quality
    B
    maintenance
    A native TypeScript AI orchestration engine and MCP server that coordinates autonomous sub-agents for complex coding tasks using DAG-based parallel execution and multi-agent consensus validation, with a real-time web dashboard.
    1
    -
  • F
    license
    Not graded
    quality
    C
    maintenance
    A production-oriented MCP runtime providing orchestration tools for IDE and agent integrations, including task management, diagnostics, and token usage reporting.
    -