riotplan-mcp-http
An HTTP-based Model Context Protocol server for RiotPlan that exposes comprehensive planning tools, resources, and prompts for AI agents. It enables full plan lifecycle management, including ideation, shaping, execution tracking, and retrospective generation.
README
@kjerneverk/riotplan-mcp-http
HTTP MCP server for RiotPlan.
This package is the network-facing surface of RiotPlan. It exposes every plan
operation as an MCP tool, resource, or prompt over HTTP using Hono and the
MCP SDK's StreamableHTTPTransport. Clients like Cursor, VS Code extensions,
and any MCP-compatible agent connect here.
What lives here
Server (src/server-hono.ts)
The Hono application that wires up MCP transport, session management, cloud
sync, RBAC authentication, plan download/upload routes, and the /health
endpoint. This is the main runtime entry point.
Tools (src/tools/)
Every riotplan_* MCP tool definition. Each file exports a tool object with
a name, Zod schema, description, and execute function. Tools cover the full
plan lifecycle:
- idea -- create plans, add notes/constraints/questions/evidence/narrative
- shaping -- start shaping, add approaches, compare, select
- build -- prepare caller-side generation instructions from plan artifacts
- build-write -- validate and persist generated plan artifacts and steps
- step -- start, complete, add, remove, move steps
- status -- read plan status
- transition -- move between lifecycle stages
- history -- checkpoints and timeline
- catalyst -- manage catalyst associations
- evidence -- structured evidence writer
- reflect -- step reflections
- retrospective -- generate plan retrospectives
- context -- read plan context for LLM consumption
- project -- bind plans to projects, resolve project context
- switch -- list plans, switch active plan, rename, delete
- generate -- server-side AI plan generation (legacy)
- validate -- plan validation
Resources (src/resources/)
MCP resource handlers for read-only access to plan data (plan metadata, status, steps, individual step content, idea, shaping, evidence, timeline, checkpoints, artifacts, prompts).
Prompts (src/prompts/)
MCP prompt templates for guided workflows (create plan, explore idea, shape approach, develop plan, execute step, execute plan, track progress, generate retrospective).
Session (src/session/)
Session management for multi-connection MCP server operation.
Other
rbac.ts-- role-based access control engine (API key auth, user/role lookup, route-level enforcement).bin-http.ts-- CLI entry point for starting the HTTP server.heartbeat.ts-- health/liveness utilities.types.ts-- MCP-specific type definitions (McpTool, ToolResult, ToolExecutionContext, resource types, prompt types).uri.ts--riotplan://URI parser.
Dependencies
| Package | Role |
|---|---|
@kjerneverk/riotplan |
Plan operations, types, AI artifact loading, config, status generation, step mutations, reflection writer, plan loader, plan categories |
@kjerneverk/riotplan-core |
Core service composition (lifecycle, status, idea, build helpers) -- used by a subset of tools |
@kjerneverk/riotplan-format |
SQLite provider for direct plan file/step/timeline access |
The dependency on @kjerneverk/riotplan is currently broad -- tools import
from subpaths like @kjerneverk/riotplan/ai/artifacts and
@kjerneverk/riotplan/config. A future goal is to narrow this so the MCP
server depends only on well-defined service interfaces rather than reaching
into riotplan internals.
Development
During development, use npm link to resolve sibling packages:
cd ../riotplan && npm link
cd ../riotplan-core && npm link
cd ../riotplan-mcp-http && npm link @kjerneverk/riotplan @kjerneverk/riotplan-core @kjerneverk/riotplan-format
Status
Extraction in progress. Source code is real (copied from riotplan/src/mcp/
with imports rewritten to use package paths). The identical source still
exists in riotplan/src/mcp/ and is tested through the riotplan test
suite. Standalone build, tests, and npm publishing are not yet configured.
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。