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
- Check the provider dependency through
ai-specs doctoror the recipe configuration flow. - Reuse a compatible
jinnafound onPATH. - If it is missing or incompatible, run
ai-specs configure-recipesinteractively and review the GitHub Release plan before accepting installation. - The installer selects the host target, verifies the provider archive with
SHA256SUMS, validates archive contents, checksjinna version, and publishes atomically. - Run
ai-specs syncafter installation so the generated MCP configuration uses eitherjinnaor 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 healthjinna 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 versionsucceeds. - 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
test-skill
Test skill for fixture purposes.
obsidian-markdown
Fixture skill for obsidian-markdown.
obsidian-cli
Fixture skill for obsidian-cli.
obsidian-bases
Fixture skill for obsidian-bases.
json-canvas
Fixture skill for json-canvas.
defuddle
Fixture skill for defuddle.