Model Context Protocol — 13 servers, ~180 tools, AI-native development
Last updated: 2026-05-22 · Hermes Agent · DeepSeek V4 Flash
📡 The Stack #
| # | Server | Tools | Area | Link |
|---|---|---|---|---|
| 1 | Firecrawl | ~24 | Web scraping, search, crawl, extraction, monitoring | firecrawl.dev |
| 2 | GitHub | ~30+ | PRs, issues, repos, code review, CI, releases | MCP GitHub Server |
| 3 | HuggingFace | ~20 | Hub search, Spaces, papers, datasets, HF Jobs | huggingface.co/mcp |
| 4 | Kaggle | ~50+ | Competitions, datasets, models, notebooks, leaderboards | kaggle.com/docs/mcp |
| 5 | W&B (Weights & Biases) | ~15 | Run tracking, traces, artifacts, reports, evals | github.com/wandb/wandb-mcp-server |
| 6 | arXiv | ~12 | Paper search, download, citation graph, alerts | github.com/blazickjp/arxiv-mcp-server |
| 7 | Browser Use | ~13 | Headless browser automation, screenshots, interaction | github.com/browser-use/browser-use |
| 8 | Filesystem | ~13 | File read/write, search, directory tree, metadata | MCP Filesystem Server |
| 9 | Context7 | ~2 | Library documentation query (Next.js, React, etc.) | mcp.context7.com |
| 10 | Git | ~10 | Local git operations (diff, log, branch, commit) | mcp-server-git |
| 11 | Time | ~2 | Timezone conversion, current time | MCP Time Server |
| 12 | Sequential Thinking | ~1 | Structured multi-step reasoning | @modelcontextprotocol/server-sequential-thinking |
| 13 | Supabase | ~? | Database, auth, storage via MCP | supabase.com/docs/guides/platform/mcp |
🔥 Firecrawl
crawl → scrape → search → map → extract → agent
🌐 firecrawl.dev — Full-stack web intelligence:
- scrape — clean markdown from any URL, JSON schema extraction
- search — web search with snippet scraping
- crawl — multi-page extraction with depth control
- map — discover URLs on a domain
- agent — autonomous research across unknown pages
- interact — click, fill forms, navigate after scrape
- monitor — recurring diffs (markdown or JSON path tracking)
- parse — local PDF/Word/Excel → markdown
Use case: Documentation extraction, competitive research, content monitoring, data pipeline ingestion.
🐙 GitHub
repos → issues → PRs → code review → actions → releases
🐙 MCP GitHub Server — Full GitHub lifecycle:
- Repositories: clone, create, fork, manage remotes
- Pull requests: diff review, inline comments, merge
- Issues: create, triage, label, assign, search
- Actions: workflow status, logs
- Releases, gists, discussions, projects
- Multi-toolset architecture: granular permission scoping
Use case: PR code review automation, issue triage, release management.
🤗 HuggingFace
hub → spaces → papers → datasets → jobs
🤗 huggingface.co/mcp — AI/ML ecosystem:
- Hub search — models, datasets, Spaces by query/filter
- Repo details — overview, dataset structure, preview rows
- Space search — semantic search, MCP-enabled Spaces, dynamic invocation
- Paper search — ML research papers on HF hub
- HF Jobs — remote compute (Python/UV/Docker) for batch inference, evaluation
- Doc search — HF library documentation
- Duplicate Space — copy Spaces (free CPU/GPU tier)
Use case: Model discovery, dataset exploration, batch inference on remote compute.
🏆 Kaggle
competitions → datasets → models → notebooks → leaderboards
🏆 kaggle.com/docs/mcp — Data science & ML competitions:
- Competitions — search, metadata, data files, submissions, leaderboards
- Datasets — search, info, file listing, metadata update, upload
- Models — list, create, update, variations, version files
- Notebooks — create session, save+run, output download, status
- Forums — topics, comments, write-ups, hackathon tracks
- Benchmarks — create tasks, leaderboards
Use case: Competition analysis, dataset ingestion, model versioning, write-up research.
📊 W&B (Weights & Biases)
runs → artifacts → traces → reports → registry
📊 github.com/wandb/wandb-mcp-server — ML experiment & LLM observability:
- Runs: query, filter, sort, history sampling, diagnosis
- Artifacts: list versions, compare diffs, lineage
- Traces (Weave): LLM call tracing, eval summary, schema inference
- Reports: create markdown reports with native chart panels
- Registry: list entities, registries, collections, versions
- Analysis: log computed data, diagnose run health
Use case: Experiment tracking, LLM observability, eval dashboards, model registry.
📄 arXiv
search → abstract → download → read → citation graph → alerts
📄 github.com/blazickjp/arxiv-mcp-server — Academic research:
- Search — advanced query syntax, field filters, category scoping
- Abstract — quick relevance check before download
- Download → Read — full paper in markdown
- Citation graph — Semantic Scholar references/citations
- Semantic search — similarity search across downloaded library
- Watch topics — standing alerts for new papers
Use case: Literature review, citation tracking, research monitoring.
🌐 Browser Use
navigate → click → type → extract → screenshot → scroll
🌐 github.com/browser-use/browser-use — Headless browser with full interaction:
- Multi-tab management (list, switch, close tabs)
- Element interaction via index or coordinates
- HTML extraction, content query, screenshots
- Session management (list, close sessions)
- Agent mode for complex multi-step tasks
Use case: Dynamic SPA content, login-required pages, form automation.
📁 Filesystem
read → write → search → tree → edit → move
📁 MCP Filesystem Server — Local file operations:
- Read/write files, directory listing with sizes
- Recursive tree view, file search with glob patterns
- Line-based edit with diff output, file move/rename
- Media file read (images, audio as base64)
Use case: Code analysis, file manipulation, project exploration.
📚 Context7
resolve library → query docs
📚 mcp.context7.com — Up-to-date library documentation:
- Resolve Library ID — map package name to Context7 ID
- Query Docs — retrieve current docs + code examples with versioning
Use case: Framework documentation queries (Next.js, React, Tailwind, etc.).
🔗 Git
status → diff → log → branch → commit → add
🔗 mcp-server-git — Local git operations:
- Status, staged/unstaged diffs, commit logs
- Branch creation, checkout, listing
- Commit with message, add files, reset staged
- Diff between branches/commits
Use case: Local version control within agent workflow.
🕐 Time
current time → convert time
🕐 MCP Time Server — Timezone utilities:
- Get current time in any IANA timezone
- Convert between timezones
🧠 Sequential Thinking
structured reasoning
🧠 @modelcontextprotocol/server-sequential-thinking — Meta-cognitive reasoning tool:
- Dynamic, reflective problem-solving
- Branch, revise, backtrack through thoughts
- Hypothesis generation and verification
Use case: Complex problem decomposition, debugging, planning.
⚡ Supabase
database → auth → storage → realtime
⚡ supabase.com/docs/guides/platform/mcp — Backend-as-a-service via MCP:
- PostgreSQL queries, table management
- Authentication, user management
- File storage
- Realtime subscriptions
📊 By the Numbers #
Total MCP Servers : 13
Total Tools : ~180
Primary Agent : DeepSeek V4 Flash
Host : macOS (Apple Silicon)
CLI : Hermes Agent → OpenCode Go
🧠 Why This Stack #
This isn't random — each server fills a role:
Research : arXiv + HuggingFace Papers + Firecrawl Search
Development : GitHub + Git + Filesystem + Context7
Data/ML : Kaggle + HuggingFace Hub + W&B
Automation : Browser Use + Firecrawl Agent + HF Jobs
Reasoning : Sequential Thinking + Time
Backend : Supabase
Tip: Stack evolves. This is a snapshot; new servers get added as workflows demand.
Part of my dev workflow — Enes Demir