Fake Pennylane MCP Server
A fake MCP server that emulates the Pennylane MCP experience for local development, testing, demos, and CI pipelines, providing deterministic fixtures and authentication modes without needing a real Pennylane account.
README
Fake Pennylane MCP Server
An open-source fake server that emulates the Pennylane MCP experience for local development, automated tests, demos, and CI pipelines — without requiring a real Pennylane account or real accounting data.
Status: working early implementation. The repository now includes a runnable Streamable HTTP MCP server, deterministic fixtures, scenario packs, static bearer auth, and a fake OAuth flow for callback/integration testing.
Why this project exists
Pennylane MCP is useful, but it is hard to validate an integration when you do not have:
- a real Pennylane account,
- a portfolio of test companies and customers,
- deterministic accounting fixtures,
- a safe environment for demos and CI.
This project solves that gap by providing a predictable, developer-friendly test double for a Pennylane-like MCP server.
Current capabilities
The current implementation provides:
- a Streamable HTTP MCP server,
- deterministic fixture loading from JSON,
- scenario-pack selection,
- optional static bearer token authentication,
- a fake OAuth authorization-code flow for end-to-end callback testing,
- automated local tests for runtime, scenarios, and auth.
Exposed tools
get_contextget_companylist_companieslist_customersget_customerlist_customer_invoices
Quickstart
Requirements
- Python 3.11+
uv
Install dependencies
cd /home/leyriel/dev/fake-pennylane-mcp-server
uv sync
Run the server
Default fixture dataset, no auth:
uv run python -m fake_pennylane_mcp_server --host 127.0.0.1 --port 8000
With a scenario pack:
uv run python -m fake_pennylane_mcp_server \
--host 127.0.0.1 \
--port 8000 \
--scenario unpaid_invoice
With static bearer auth:
uv run python -m fake_pennylane_mcp_server \
--host 127.0.0.1 \
--port 8000 \
--auth-mode static_bearer \
--bearer-token test-secret-token
With fake OAuth flow:
uv run python -m fake_pennylane_mcp_server \
--host 127.0.0.1 \
--port 8000 \
--auth-mode fake_oauth
Run with Docker Compose
Build and start the default server:
docker compose up --build
Run detached:
docker compose up --build -d
Stop it:
docker compose down
If port 8000 is already used on your machine, override the published host port:
PUBLISHED_PORT=8080 docker compose up --build -d
Use a scenario pack:
SCENARIO_NAME=unpaid_invoice docker compose up --build
Use static bearer auth:
AUTH_MODE=static_bearer BEARER_TOKEN=test-secret-token docker compose up --build
Use fake OAuth mode:
AUTH_MODE=fake_oauth docker compose up --build
By default Compose exposes the server on http://127.0.0.1:${PUBLISHED_PORT:-8000}/mcp and mounts ./fixtures read-only into the container.
Scenario packs
The default fixture file currently exposes these scenario packs:
defaultcustomer_foundcustomer_not_foundunpaid_invoicemulti_company
Each scenario narrows the deterministic dataset to a stable business situation.
Authentication modes
none
Default mode. No Authorization header is required.
static_bearer
Requires a bearer token on MCP HTTP requests.
Example header:
Authorization: Bearer ***
fake_oauth
Exposes a minimal fake OAuth authorization server alongside the MCP resource server.
Included endpoints:
/.well-known/oauth-authorization-server/.well-known/oauth-protected-resource/mcp/authorize/token
Current fake client contract:
client_id:fake-public-client- token endpoint auth method:
none - redirect URIs:
http://127.0.0.1:9999/callbackhttp://localhost:9999/callback
- scope:
mcp:access
This flow is meant for local integration tests and fake callback simulations — not as a production OAuth server.
Compatibility and status tracking
See:
docs/compatibility-matrix.mddocs/README.md
Example clients
See:
examples/basic_client.pyexamples/bearer_client.pyexamples/README.md
Running tests
uv run pytest -q
Development workflow
Typical local loop:
- update fixtures / scenarios / auth behavior,
- run
uv run pytest -q, - run one of the example clients against a local server,
- commit the change.
Project structure
.
├── docs/
├── examples/
├── fixtures/
├── src/
├── tests/
├── CONTRIBUTING.md
├── Dockerfile
├── docker-compose.yml
├── LICENSE
├── README.md
└── pyproject.toml
Roadmap
Bootstrap
- [x] Create Plane project
- [x] Create public GitHub repository
- [x] Publish initial detailed README
- [x] Prepare repo structure for implementation handoff
Implementation milestones
- [x] Implement fake Streamable HTTP MCP runtime
- [x] Add deterministic fixture loading
- [x] Add fake customer / invoice queries
- [x] Add scenario packs
- [x] Add auth modes for integration testing
- [x] Add runnable example clients
- [x] Add fake OAuth flow for end-to-end callback testing
- [x] Publish a compatibility matrix
- [x] Add Docker and Docker Compose local runtime
Next useful milestones
- [ ] add GitHub Actions CI once a token with
workflowscope is available - [ ] enrich compatibility with more Pennylane-like tools
- [ ] add more domain fixture packs
- [ ] add standalone OAuth example client
- [ ] document additional integration recipes
Open-source direction
The intent is for this project to be genuinely useful to a broader audience, not just a one-off internal spike.
That means we optimize for:
- clarity of documentation,
- ease of local setup,
- stable deterministic fixtures,
- transparent scope,
- simple contribution paths.
Project origin
This repository was spun out from work around a MissionGuard integration effort, where a Pennylane MCP replacement strategy needed a realistic test environment without depending on a real Pennylane tenant.
License
MIT.
Disclaimer
This project is not affiliated with or endorsed by Pennylane. It is an independent open-source testing utility intended to emulate part of a Pennylane-like MCP integration surface for development and testing purposes only.
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。