thinker-mcp
MCP server for thinker-ci, exposing tickets, CI/CD pipelines, project management data, and AI-agent sessions as tools and resources for Claude Code and other MCP clients.
README
thinker-mcp
MCP server for the thinker-ci platform. Exposes tickets, CI/CD pipelines, project-management data, and AI-agent sessions as MCP tools and resources so that Claude Code (or any other MCP-capable client) can interact with the platform directly from the assistant.
Architecture
Claude Code / Cursor / other MCP client
│ stdio / SSE
▼
thinker-mcp ← this repo
┌──────────────────────────────────┐
│ tools/tickets.py (file-system) │─── reads/writes thinker-blueprint/
│ tools/pipelines.py │─── thinker-console REST API
│ tools/pm.py │─── thinker-console REST API
│ tools/agents.py │─── thinker-agent REST API
│ resources/schema.py │─── thinker-console /api/schema/
│ resources/docs.py │─── thinker-blueprint/ (local)
└──────────────────────────────────┘
│ │
thinker-console thinker-agent
(Django REST API) (LLM agent API)
│
thinker-blueprint
(git repo on disk)
thinker-console is the central Django application that owns projects, pipelines, runs, PM objects (epics, stories, tasks, sprints, milestones), and user/tenant data.
thinker-agent is a lightweight FastAPI service that manages stateful LLM sessions. It proxies requests to Anthropic, OpenAI, Google, and AWS Bedrock.
thinker-blueprint is a git repository of Markdown files that serves as the project's file-system ticketing system. The tickets/ tree is readable and writable by thinker-mcp without going through an API.
Installation
From source (development)
git clone git@github.com:thinker-ci/thinker-mcp.git
cd thinker-mcp
pip install -e .
Via uvx (no checkout required)
uvx thinker-mcp
Configuration
Copy .env.example to .env and fill in the values:
cp .env.example .env
$EDITOR .env
| Variable | Default | Description |
|---|---|---|
THINKER_CONSOLE_URL |
http://localhost:8000 |
Base URL of the thinker-console Django API |
THINKER_CONSOLE_TOKEN |
(required) | API token for authenticating to thinker-console (Token <value>) |
THINKER_AGENT_URL |
http://localhost:8080 |
Base URL of the thinker-agent FastAPI service |
THINKER_AGENT_TOKEN |
(required) | Bearer token for authenticating to thinker-agent |
THINKER_BLUEPRINT_PATH |
/data/thinker-blueprint |
Absolute path to a local checkout of thinker-blueprint |
Tools
Ticket tools (file-system)
These tools read from and write to the tickets/ tree inside a local thinker-blueprint checkout. No network calls are made.
| Tool | Parameters | Description |
|---|---|---|
tickets_list |
status, type, sprint, assignee |
List tickets with optional filters |
ticket_get |
id |
Get full markdown content of a ticket (e.g. STORY-003) |
ticket_update_status |
id, status |
Update the status: frontmatter field |
ticket_create |
type, title, description, epic_id, priority, sprint |
Create a new ticket file |
Pipeline tools (thinker-console)
| Tool | Parameters | Description |
|---|---|---|
pipelines_list |
project_slug |
List pipelines for a project |
pipeline_trigger |
pipeline_id, branch, commit_sha? |
Trigger a pipeline run |
run_get |
run_id |
Get status and metadata for a run |
run_logs |
job_id |
Fetch raw logs from a job |
run_cancel |
run_id |
Cancel a running pipeline run |
runners_list |
— | List all CI runners and their status |
Project-management tools (thinker-console)
| Tool | Parameters | Description |
|---|---|---|
epics_list |
status?, milestone? |
List epics |
stories_list |
epic_id?, sprint_id?, status? |
List stories |
tasks_list |
story_id?, assignee?, status? |
List tasks |
story_update |
story_id, status?, story_points? |
Update a story |
task_create |
title, story_id, task_type?, estimate? |
Create a task |
sprint_board |
— | Current sprint kanban board |
sprints_list |
— | List all sprints |
milestones_list |
— | List all milestones |
Agent tools (thinker-agent)
| Tool | Parameters | Description |
|---|---|---|
agent_session_create |
provider, model_id, system_prompt? |
Create a new LLM session |
agent_message_send |
session_id, message |
Send a message; returns the assistant response |
agent_sessions_list |
— | List all active sessions |
Resources
| URI | MIME type | Description |
|---|---|---|
thinker://openapi/schema |
application/json |
Live OpenAPI 3.x schema from thinker-console |
thinker://docs/architecture |
text/markdown |
Concatenated architecture/*.md from thinker-blueprint |
thinker://tickets/<stem> |
text/markdown |
Individual ticket file (e.g. thinker://tickets/STORY-003-tenant-model) |
Using with Claude Code
Add to your project's .claude.json (or ~/.claude.json for global access):
{
"mcpServers": {
"thinker": {
"command": "thinker-mcp",
"env": {
"THINKER_CONSOLE_URL": "https://console.thinker.ci",
"THINKER_CONSOLE_TOKEN": "your-token-here",
"THINKER_AGENT_URL": "https://agent.thinker.ci",
"THINKER_AGENT_TOKEN": "your-agent-token-here",
"THINKER_BLUEPRINT_PATH": "/home/you/thinker-blueprint"
}
}
}
}
Or use a .env file in your working directory instead of inline env keys.
Restart Claude Code after editing the config. Run /mcp to verify the server is connected and tools are listed.
Using with other MCP clients
Any client that supports the MCP stdio transport can connect by spawning:
thinker-mcp
The server reads STDIN and writes STDOUT using the JSON-RPC framing defined by the MCP protocol. Set the environment variables above before launching.
Running in Docker
docker build -t thinker-mcp .
docker run --rm -i \
-e THINKER_CONSOLE_URL=http://console:8000 \
-e THINKER_CONSOLE_TOKEN=secret \
-e THINKER_AGENT_URL=http://agent:8080 \
-e THINKER_AGENT_TOKEN=secret \
-v /data/thinker-blueprint:/data/thinker-blueprint:ro \
thinker-mcp
Development
# Install in editable mode with dev extras (add [dev] if you add them to pyproject.toml)
pip install -e .
# Lint
ruff check src/
# Format
ruff format src/
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。