Swistaczek MCP

MCP运维云端

提供方 swistaczek

运维云端 未标注 无需鉴权 社区 / 第三方 Claude

一个基于Python 3.13和FastMCP的个人服务器集合,用于管理多种功能和服务。

一句话选型结论

按资料字段整理

Swistaczek MCP 属于运维云端类 MCP 服务器,传输方式未标注,无需鉴权,可直接接入 Claude,没有现成安装命令,需按官方文档接入。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类运维云端 传输方式未标注 鉴权无需鉴权 部署方式未标注
GitHub 星标 0 仓库热度,无公开仓库时显示未标注
工具数 19 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

运维云端
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量19暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

Python服务器Python本地部署FastMCP多功能服务API集成公共数据

安装与配置

未标注 无需鉴权
上游未给出传输方式
  1. 1
    准备运行环境

    按官方文档确认运行环境(常见为 Node.js 或 Python),装好后再执行安装命令。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后客户端会列出约 19 个工具,随便调一个只读工具确认能返回结果。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
云资源查询

查实例、存储、账单等云上资源状态。

部署与发布

把部署、重启、回滚这类操作接进对话。

监控与告警

拉取监控指标,异常时让模型帮你判断。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 资料标注 19 个工具,覆盖面较广,能满足多种调用。
  • 已适配 Claude,换客户端不用重新研究配置。
  • 资料标注无需鉴权,装好就能连。

先想清楚的地方

  • 没有现成安装命令,需要按官方文档自己填启动命令和参数。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在运维云端场景里让模型直接调用外部能力的人。
  • 已经在用 Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Swistaczek MCP当前 19 未标注 未标注 无 2026-05-17
lobehub 未标注 stdio · 本地进程 本地部署 有 2026-04-11
TrendRadar 未标注 stdio · 本地进程 本地部署 有 2026-04-13
Ui Tars Desktop 8a825770 未标注 stdio · 本地进程 本地部署 有 2026-05-17
ruflo 未标注 stdio · 本地进程 本地部署 有 2026-04-13
FastGPT 未标注 stdio · 本地进程 本地部署 有 2026-04-11

服务器档案

运维云端
字段来自项目仓库与公开索引,中文说明由本站标注
提供方swistaczek上游标注的维护方 swistaczek
分类运维云端本站按上游分类展示
连接方式未标注上游未给出传输方式
鉴权方式无需鉴权远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型未标注上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

15.1 KB
来自项目仓库与上游目录的原文,未做删改

MCP Servers

Personal MCP servers collection — Python 3.13 + FastMCP, managed with mise and uv.

Requirements

  • mise (brew install mise) — provisions Python 3.13 and uv automatically from .mise.toml.
  • That's it. mise picks up uv from [tools] in .mise.toml; you don't need to install python or uv yourself.

Quick Start```bash

git clone https://github.com/swistaczek/mcp.git
cd mcp
mise install # provisions Python 3.13 + uv from .mise.toml
mise run install # uv sync — installs all project deps from uv.lock
mise run test # 141 tests, ~3s, no network


To register the servers with Claude Code (user-scope, one-shot):```bash
mise run claude-install                   # registers key-less servers
GEMINI_API_KEY=sk-... mise run claude-install-gemini   # registers Gemini servers

Then restart Claude Code and the servers appear under their slug names (czynaczas, gazetki, tablica, exif_extractor, domains, krs, poznan_events, plus gemini_image_descriptions and plate_recognition if you ran the second task). Remove with mise run claude-uninstall.

Configuration

API keys

gemini_image_descriptions and plate_recognition need GEMINI_API_KEY (free at AI Studio). Two ways to provide it:

  1. Claude Code settings — add to .claude/settings.local.json (gitignored):json { "env": { "GEMINI_API_KEY": "your-key-here" } }
  2. Environment — cp .env.example .env, edit, and use mise run claude-install-gemini which forwards the var into the registration via claude mcp add -e.

Two install paths

Servers can be registered two ways and both work:

  • Direct python (recommended) — what mise run claude-install does. Points Claude Code at .venv/bin/python .py. Fast cold-start, no fastmcp.json indirection, no uv shell-out.
  • fastmcp.json — what .mcp.json uses (read by Claude Code on session start when this repo's directory is its cwd). Goes through uv run --project . fastmcp run .fastmcp.json. Requires uv on PATH and the project's cwd to be this repo.

If you bind the repo as a workspace in Claude Code, .mcp.json is the easy path. For user-scope (always-on) registration outside this repo, use mise run claude-install.

Available Servers

Czy Na Czas (czynaczas.py)

Polish public-transport realtime data via czynaczas.pl — vehicle positions, delays, ETAs.
- Cities: poznan, warsaw, krakow, wroclaw, lodz
- Tools:
- list_supported_cities — list the five cities
- find_stops(city, query) — diacritic-insensitive multi-token search across the 3000+ stop list per city ("Dabrowskiego" matches "Dąbrowskiego")
- get_trip(city, trip_id) — route polyline + scheduled stops
- get_vehicle(city, vehicle_id) — fleet metadata (model, depot, A/C, low-floor)
- get_departures(city, stop_id, destination_filter?, line_filter?) — headline tool. Snapshots all live vehicles via Socket.IO, filters by destination headsign + line, and computes ETA from each candidate's scheduled arrival plus realtime delay.
- Usage: "Kiedy najbliższy tramwaj 16 z Polna na Ogrody?" → find_stops → get_departures(stop_id, destination_filter="Ogrody").

Gazetki (gazetki.py)

Current promotional flyer PDFs from Polish supermarket chains — Lidl and Biedronka.
- Tools:
- list_flyers(chain) — lidl / biedronka / all
- download_flyer(chain, flyer_id) — streams native PDFs (Lidl) or assembles from per-page images (Biedronka)
- get_current_flyers(chain?) — convenience: headline current flyer per chain
- Cache: ~/.cache/mcp-gazetki/// (override with GAZETKI_CACHE_DIR).

Tablica Rejestracyjna PL (tablica.py)

Polish license-plate violation reporting via tablica-rejestracyjna.pl.
- Tools: fetch_comments, submit_complaint (with image upload + HEIC→JPEG)

EXIF Metadata Extractor (exif_extractor.py)

Extract EXIF + GPS from images and reverse-geocode to street addresses.
- Tool: analyze_image_metadata (up to 50 images, PNG/JPEG/HEIC, Nominatim geocoding)

Plate Recognition (plate_recognition.py)

Identify license plates and traffic violations in photos via Gemini Vision.
- Tool: recognize_plates (multi-vehicle, pedestrian-perspective reasoning)
- Setup: requires GEMINI_API_KEY

Image Descriptions (gemini_image_descriptions.py)

Generate alt text and accessible descriptions for images and GIFs via Gemini.
- Tool: generate_image_descriptions (batch up to 20 images, GIF support via FFmpeg)
- Setup: requires GEMINI_API_KEY

KRS Poland (krs.py)

Polish National Court Register (KRS) Open API — fetch the current or full extract for any registered entity.
- Tools:
- get_krs_current_extract(krs, rejestr?) — odpis aktualny: name, NIP, REGON, formaPrawna, address, board, capital
- get_krs_full_extract(krs, rejestr?) — odpis pełny: full historical record (past names, board changes, all entries)
- Inputs: KRS is 10 digits starting with 0 (e.g. 0001236495); rejestr is P (companies) or S (associations/foundations), defaults to P. A 204 response hints at trying the other register.
- Source: api-krs.ms.gov.pl/api/krs (no key required). Port of krs-poland-mcp-server.

Poznań Events (poznan_events.py)

City of Poznań events calendar (Co? Gdzie? Kiedy? — poznan.pl/mim/events/).
- Tools:
- list_events(category?, page?, pages?) — paginated event list (20 per page); optional category id filter; up to 5 pages per call
- get_event(event_ref) — full detail (title, ISO date, time, place, categories, short + long description, hero image). Accepts numeric id, slug+id, path, or full URL
- list_event_categories — catalog of category names with numeric ids (Sport=214, Muzyka=217, Sztuka=218, Teatr, Film, Książki, Dziecko, …)

Label Printer (label_printer.py)

Prints courier labels (InPost, DPD, DHL…) on a direct-thermal label printer attached to this machine. Works as an MCP server and a standalone CLI.
- Tools:
- list_thermal_printers — detects connected label printers, whether each has a CUPS queue, and its page-description language
- print_label(file_path, printer?, label_size?, copies?, darkness?, speed?, threshold?, bold?, dry_run?) — renders and prints a PDF or image; creates the CUPS queue on first use
- print_qr_code(data, printer?, module_dots?, error_correction?, caption?) — prints a QR code for any URL or text
- get_printer_status(printer?) — queue state and pending jobs
- cancel_print_jobs(printer?, job_id?) — clear the queue
- CLI: uv run python label_printer.py detect · status · install · render · print · qr · cancel (add --json for machine-readable output)
- Two transports: CUPS-backed label printers (Zebra EPL2/ZPL, Dymo) go through lp; cheap ESC/POS receipt printers, which no bundled CUPS driver can drive, are handled directly over libusb by escpos_printer.py. Both are listed together and picked automatically.
- Rendering: thermal heads are 1-bit, so the label is rasterised at 4× resolution, box-filtered down to the exact dot grid, and hard-thresholded to pure black/white. Letting the print system dither an anti-aliased page instead produces visibly faint output with barcodes that scan poorly.
- Auto-detection: with a single thermal printer attached, printer can be omitted everywhere. Zebra, Dymo, TSC, Godex, Citizen, Bixolon, Sato, Argox, Intermec, Honeywell, Brother QL, Toshiba TEC and Seiko devices are recognised.
- Requires: pdftoppm (brew install poppler) for PDF input, plus the CUPS command-line tools. Verified on macOS with a Zebra TLP2844 (USB, EPL2, 203 dpi) and 4×6" stock.
- Usage: "Print this InPost label" → print_label(file_path="~/Downloads/label.pdf"). Darker output → raise threshold (up to 254) or set bold=1.

Domain Checker (domains.py)

Batch domain registration check via WHOIS with DNS fallback and optional OVH browser-verified availability.
- Tool: check_domains (up to 50 domains, 50+ TLDs incl. .com.cn and Chinese IDN)

Google Ads Planner (google_ads_planner.py)

Keyword research, demand forecasting and budget planning against the Google Ads API (v25, REST).
- Tools:
- google_ads_keyword_metrics(keywords, geo_target_ids?, language_id?, network?, start_year_month?, end_year_month?) — "how much demand is there for these exact terms?" Historical average monthly searches, competition band + index, top-of-page bid range, and the month-by-month volume series. Up to 10,000 keywords per call.
- google_ads_keyword_ideas(seed_keywords? · seed_url? · seed_site?, geo_target_ids?, language_id?, network?, limit?) — "what else should we be targeting?" Expands one seed kind (max 20 seed keywords, or a single URL/site) into related keywords, each scored with an opportunity_score (volume ÷ competition), plus ready-made top_by_volume, top_opportunities and easy_targets (LOW/MEDIUM competition) buckets.
- google_ads_forecast_budget(keywords, match_type?, max_cpc_bid, daily_budget?, start_date?, end_date?, …) — "what would this cost and what would we get?" Forward-looking forecast for a hypothetical campaign: clicks, cost, average CPC, conversions, average CPA.
- google_ads_budget_curve(keywords, match_type?, bids?, …) — "where does spending more stop paying off?" Sweeps the forecast across a bid ladder (default 0.25 … 8.00, max 10 points), computes marginal clicks / marginal cost / marginal cost-per-click between consecutive points, and picks the best_efficiency_bid.
- google_ads_seasonality(keywords, months_back?, geo_target_ids?, language_id?) — "when in the year should we spend?" Up to 48 months of history normalised into a monthly index (1.0 = average month), with peak/trough months per keyword and in aggregate, plus year-over-year change.
- Defaults: geo 2840 (USA), language 1000 (English), network GOOGLE_SEARCH. UK is 2826, Canada 2124; Spanish 1003, French 1002.
- Setup: see Google Ads Planner setup below — credentials come from 1Password, not from files.
- Caveats:
- A Basic-access-tier developer token is required. Test and Explorer tier tokens cannot call the keyword planning endpoints at all — they fail regardless of the account or the query. Apply for Basic access in your manager account's API Center.
- Keyword planning is rate-limited to 1 QPS per customer id. The server serialises and paces every request itself, so a google_ads_budget_curve over six bids takes ~7s of wall clock by design. Don't fan these tools out in parallel.
- The v25 forecast returns no impressions and no CTR — those fields were removed in v24. You get clicks, cost, average CPC, conversions and average CPA only; impression share and click-through rate cannot be derived from it.
- Absent numbers mean "no data", not zero. Bid and competition fields are frequently missing for long-tail keywords and come back as null — don't read them as $0.00 or as zero competition.
- Source: googleads.googleapis.com REST endpoints under customers/{id}:generateKeywordHistoricalMetrics / :generateKeywordIdeas / :generateKeywordForecastMetrics.

Google Ads Planner setup

The server reads four environment variables. Nothing secret is stored in this repo —
.mcp.json holds only 1Password references, and op run resolves them at launch.

Variable What it is
GOOGLE_ADS_DEVELOPER_TOKEN API token from your manager account's API Center
GOOGLE_ADS_SERVICE_ACCOUNT_KEY_B64 base64 of the GCP service-account JSON key
GOOGLE_ADS_LOGIN_CUSTOMER_ID manager (MCC) account id, digits only
GOOGLE_ADS_CUSTOMER_ID the client account actually queried, digits only
GOOGLE_ADS_API_VERSION optional, defaults to v25

1. Get a developer token (Basic tier)

Only a manager (MCC) account can issue one. In the MCC: Tools & Settings → Setup →
API Center
(). Apply for Basic access — new
tokens start on Explorer tier, which cannot call keyword planning at all, and Test tier
only works against test accounts. Approval is manual, typically 1–3 business days.

2. Create the service account

Auth is a GCP service account with no impersonation (no domain-wide delegation, no
Workspace admin needed). Create a JSON key, then add the service account's email as a
user on the Ads account under Admin → Access and security. Encode the key:```bash
base64 -i service-account.json | tr -d '\n'


### 3. Store the credentials in 1Password

Create one API Credential item — the reference implementation lives in the `Startupkit`
vault as **Google Ads MCP**:

| Field | Contents |
| --- | --- |
| `credential` | developer token |
| `ads_manager_account_id` | MCC id |
| `company_account_id` | client account id |
| `Service Account` → `SERVICE_ACCOUNT_KEY_B64` | the base64 blob from step 2 |

> **Item titles must not contain `(`, `)` or other punctuation** — `op://` secret
> references reject them with `invalid character in secret reference`. Stick to letters,
> digits, spaces and hyphens.

Verify each reference resolves before wiring it up:```bash
op read "op://Startupkit/Google Ads MCP/credential"

4. Launch

.mcp.json already wraps the server in op run, so Claude Code picks it up with no
further setup as long as you have an active op session (op signin):```json
"args": ["-c", "op run --no-masking -- uv run --with fastmcp fastmcp run google_ads_planner.fastmcp.json"],
"env": {
"GOOGLE_ADS_DEVELOPER_TOKEN": "op://Startupkit/Google Ads MCP/credential"
}


`--no-masking` is **required**, not optional. `op run` otherwise scans stdout and
replaces anything resembling a secret with `` — and for a stdio MCP server
stdout *is* the JSON-RPC channel, so masking can corrupt the protocol stream.

To point at a different vault or item, edit the `op://` references in `.mcp.json`; they
are not secrets and are safe to commit.

### Running outside Claude Code

Same mechanism, driven from a dotenv file of references:```bash
cat > .env.google-ads <<'EOF'
GOOGLE_ADS_DEVELOPER_TOKEN=op://Startupkit/Google Ads MCP/credential
GOOGLE_ADS_SERVICE_ACCOUNT_KEY_B64=op://Startupkit/Google Ads MCP/Service Account/SERVICE_ACCOUNT_KEY_B64
GOOGLE_ADS_LOGIN_CUSTOMER_ID=op://Startupkit/Google Ads MCP/ads_manager_account_id
GOOGLE_ADS_CUSTOMER_ID=op://Startupkit/Google Ads MCP/company_account_id
EOF

op run --no-masking --env-file=.env.google-ads -- mise run dev google_ads_planner.fastmcp.json

.env* is gitignored. If you genuinely cannot use 1Password, plain values in .env
work too — but never put them in .mcp.json, which is committed.

Development```bash

mise tasks # list all available tasks
mise run dev <server>.fastmcp.json # run a server in stdio mode
mise run test # quick (no integration)
mise run test-all # full (network + GEMINI_API_KEY needed)
```

To create a new server: copy czynaczas.py and czynaczas.fastmcp.json as a template, add the module name to [tool.setuptools] py-modules in pyproject.toml, and re-run mise run install. See FastMCP docs for the full pattern.

Links

  • FastMCP · mise · uv

常见问题

6 问
答案由本站依据条目字段整理,不引用上游原文
Swistaczek MCP 怎么安装?

资料没有给出安装命令,请到官方仓库或文档按说明接入;这类条目多数是生态应用,或尚未发布可安装包。

Swistaczek MCP 怎么在 Claude 里配置?

它的传输方式未标注,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Swistaczek MCP 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Swistaczek MCP 需要 API Key 吗?

资料标注为无需认证,装好就能连;如果实际调用时报鉴权错误,按官方文档补上凭据。

Swistaczek MCP 支持哪些客户端?

资料标注支持 Claude;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Swistaczek MCP 能提供哪些能力?

资料显示约 19 个工具;能力标签包括 Python服务器、Python、Claude。具体工具清单以官方文档为准。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。