AGENT SKILL

jinna-mcp-recipe

Install and use the local OpenProject provider MCP through ai-specs; detect the provider binary, configure OpenProject environment references, and keep local/official MCP boundaries explicit.

License 许可
MIT
最近更新
2026-11-08
Stars
★ 0
Forks
2

01 前言 Front Matter

name
jinna-mcp-recipe
license
MIT
metadata
{ "scope": "runtime", "author": "ai-specs", "version": "1.0", "auto_invoke": [ "Using the OpenProject provider MCP", "Configuring jinna-mcp-recipe", "Installing or troubleshooting the OpenProject provider" ] }
description
Install and use the local OpenProject provider MCP through ai-specs; detect the provider binary, configure OpenProject environment references, and keep local/official MCP boundaries explicit.

03 SKILL.md 正文

OpenProject provider MCP

Terminology

  • Provider: the separate Go project parada1104/jinna-provider.
  • Provider binary: the local executable jinna.
  • OpenProject: the remote Self-Hosted service configured by OPENPROJECT_BASE_URL.
  • Local MCP: jinna mcp, a stdio server that uses OpenProject APIv3.
  • Python client: reference-only material; do not require or invoke it.

Installation flow

  1. Check the provider dependency through ai-specs doctor or the recipe configuration flow.
  2. Reuse a compatible jinna found on PATH.
  3. If it is missing or incompatible, run ai-specs configure-recipes interactively and review the GitHub Release plan before accepting installation.
  4. The installer selects the host target, verifies the provider archive with SHA256SUMS, validates archive contents, checks jinna version, and publishes atomically.
  5. Run ai-specs sync after installation so the generated MCP configuration uses either jinna or the verified managed executable path.

Never assume that OpenProject itself is installed locally. Never require Go, Mise, Python, Git, GitHub CLI, or a package manager for a provider release binary.

Configuration

Keep credentials outside generated files:

OPENPROJECT_BASE_URL
OPENPROJECT_API_TOKEN
OPENPROJECT_AUTH (optional: basic or bearer)

Use the existing ai-specs environment flow and user-owned ignored environment file. Do not put token values in MCP arguments, manifests, documentation, logs, or tool payloads.

MCP boundary

Use the recipe-provided local jinna mcp server. The official OpenProject remote /mcp endpoint is a separate operator-selected option with its own server configuration and authentication. Do not automatically proxy, switch, or replay writes between these protocols.

Runtime boundary

ai-specs sync and ai-specs doctor are local materialization checks; they do not contact OpenProject. Live diagnostics are opt-in commands:

  • jinna health
  • jinna whoami

Those commands make a live request to the configured service, and no passive command runs them automatically. The recipe's 30-second MCP timeout bounds local jinna mcp startup and transport, not the remote OpenProject API SLA. Keep slow or unavailable self-hosted service behavior visible: do not hide it with automatic retries, proxy fallback, endpoint switching, or write replay.

Safe troubleshooting

  • Missing provider: configure interactively and accept the allowlisted GitHub Release plan, or follow the printed manual URL.
  • Unknown version: treat the provider as unresolved until jinna version succeeds.
  • Checksum or archive failure: do not execute the candidate; retain the previous managed version.
  • Missing OpenProject settings: complete environment configuration; provider installation itself does not contact OpenProject.
  • Non-interactive or CI run: expect detection and guidance only, never a download.

04 同仓库更多 Skills