MCP server and client in one ~8 MB binary. Connect to any stdio MCP server, expose your own tools, and run AI pipelines. 246 builtins. Zero dependencies.
v1.2.0 · ~8 MB single binary · Linux · macOS · Windows · Raspberry Pi
v1.2.0 · ~8 MB einzelne Binary · Linux · macOS · Windows · Raspberry Pi
curl -fsSL https://pipe-lang.com/install.sh | bash
A real pipeline: classify → summarize → translate → save. Run it yourself below ↓
Eine echte Pipeline: klassifizieren → zusammenfassen → übersetzen → speichern. Probier sie unten selbst aus ↓
Named agents with their own prompt and tools transfer a conversation to one another via a reserved tool call — full shared history, the pattern OpenAI's original "Swarm" popularized.
Benannte Agenten mit eigenem Prompt und Tools reichen eine Konversation per reserviertem Tool-Call weiter — geteilter Verlauf, das Pattern von OpenAIs ursprünglichem „Swarm“.
Answer questions about an image — a URL, a local file, or raw bytes — against DeepSeek's vision model. "photo.jpg" > ai_vision "What's in this?"
Fragen zu einem Bild beantworten — URL, lokale Datei oder rohe Bytes — gegen DeepSeeks Vision-Modell. "foto.jpg" > ai_vision "Was ist zu sehen?"
pipe --update checks GitHub, verifies the SHA256 checksum, and replaces the running binary atomically. pipe --version / --update-check included.
pipe --update prüft GitHub, verifiziert den SHA256-Checksum und ersetzt die laufende Binary atomar. pipe --version / --update-check inklusive.
The 6th provider runs free-tier models with zero API key at all: ai_provider "opencode" and go.
Der 6. Provider läuft mit Free-Tier-Modellen ganz ohne API-Key: ai_provider "opencode" und los.
Round 7 (CLI --sandbox missed 6 fs-write builtins) and round 8 (wiki_search bypassed the AI egress gate entirely) — found, fixed, live-verified.
Runde 7 (CLI --sandbox verfehlte 6 Schreib-Builtins) und Runde 8 (wiki_search umging den KI-Egress-Gate komplett) — gefunden, gefixt, live verifiziert.
A stale bytecode cache could resolve a builtin call to the wrong function after the builtin table changed. Now self-invalidating — no known repeat possible.
Ein veralteter Bytecode-Cache konnte einen Builtin-Aufruf nach einer Änderung der Builtin-Tabelle auf die falsche Funktion auflösen. Jetzt selbst-invalidierend — keine Wiederholung mehr möglich.
No install. No signup. Just type Pipe code and run.
Keine Installation. Keine Anmeldung. Einfach tippen und ausführen.
LLMs with file access, network, and exec are a liability. You need sandboxing at the language level — not afterthought middleware.
LLMs mit Dateizugriff, Netzwerk und exec sind ein Risiko. Du brauchst Sandboxing auf Sprachebene — kein nachträgliches Middleware-Gefrickel.
Sequential API calls turn a 1-second pipeline into a 10-second bottleneck. Parallelism shouldn't require asyncio.gather() boilerplate.
Sequentielle API-Calls machen aus einer 1-Sekunden-Pipeline einen 10-Sekunden-Flaschenhals. Parallelismus sollte kein asyncio.gather()-Boilerplate brauchen.
Switching from OpenAI to DeepSeek means rewriting your SDK code. Provider changes should be one line — not a refactor.
Von OpenAI zu DeepSeek wechseln heißt SDK-Code umschreiben. Provider-Wechsel sollten eine Zeile sein — kein Refactor.
Pipe fixes this at the language level. Pipe löst das auf Sprachebene.
Read server logs, filter critical entries, summarize findings with AI, translate to German, and save — 8 lines, measured in benchmarks/python-vs-pipe. No intermediate files. No Python script.
Server-Logs einlesen, kritische Einträge filtern, per KI zusammenfassen, ins Deutsche übersetzen und speichern — 8 Zeilen, gemessen in benchmarks/python-vs-pipe. Keine Zwischendateien. Kein Python-Skript.
ai_provider "deepseek" logs: read_lines "data/incident.log" errors: filter logs (fn line: (len (split line "ERROR")) > 1) joined: join errors "\n" summary: summarize joined german: translate summary "de" write_file "incident_report.md" ("# Incident Report\n## Deutsch\n" ++ german ++ "\n") print german
Vectorize documents, find matches by meaning — not keywords. Built-in embed, nearest, cosine_sim. Works with every provider: OpenAI, DeepSeek, Anthropic, Ollama, OpenRouter. No vector DB setup.
Dokumente vektorisieren, Treffer nach Bedeutung finden — nicht nach Stichwörtern. Eingebaute embed, nearest, cosine_sim. Funktioniert mit jedem Provider: OpenAI, DeepSeek, Anthropic, Ollama, OpenRouter. Keine Vektor-DB.
ai_provider "deepseek" docs: [read_file "data/docs/database.txt"] push docs (read_file "data/docs/caching.txt") push docs (read_file "data/docs/api.txt") push docs (read_file "data/docs/deployment.txt") vectors: embed_batch docs question: "How do we rate-limit API requests?" q_vec: embed question top: nearest q_vec vectors 3 context: "" for idx in top context: context ++ (at docs idx) ++ "\n---\n" answer: ask ("Context:\n" ++ context ++ "\nQuestion: " ++ question) print answer
Define a tool, register it with the LLM, and let the model call it autonomously. Sandbox profiles lock down exec, write_file, and network access. The same code swaps between OpenAI, DeepSeek, OpenRouter, and Ollama with one line.
Ein Tool definieren, beim LLM registrieren und das Modell autonom aufrufen lassen. Sandbox-Profile sperren exec, write_file und Netzwerkzugriff. Derselbe Code wechselt mit einer Zeile zwischen OpenAI, DeepSeek, OpenRouter und Ollama.
-- Declare a sandbox: temp files only, network ok, no exec sandbox_profile "agent" {fs: "temp-only", network: true, exec: false, ai: true} set_sandbox "agent" fn get_weather city match city | "Berlin" -> "22°C, sunny" | "London" -> "15°C, rainy" | _ -> city ++ ": no data" ai_tool "get_weather" "Get current weather for a city" {city: "City name"} get_weather ai_with_tools "You are a weather assistant." "What's the weather in Berlin and London?" > print
Pipe is the first language with built-in MCP — both server and client. Connect to any stdio MCP server off npm/uvx, expose your own tools to Claude Desktop, and use everything together in ai_with_tools. Zero dependencies, pure Go stdlib.
Pipe ist die erste Sprache mit eingebautem MCP — Server und Client. Verbinde dich mit jedem stdio-MCP-Server per npm/uvx, stelle eigene Tools für Claude Desktop bereit und nutze alles zusammen in ai_with_tools. Zero Dependencies, reines Go.
mcp_use_stdio "npx" "-y" "@modelcontextprotocol/server-github" {GITHUB_TOKEN: (env "GITHUB_TOKEN")} mcp_use_stdio "npx" "-y" "@modelcontextprotocol/server-filesystem" "/tmp" ai_with_tools "You have GitHub + filesystem tools." "Check pipe's open issues, list /tmp files, save a summary."
Built-in Discord webhook and Telegram Bot modules. Send rich embeds with AI code reviews from CI, post notifications, or build custom chat bots — pure Pipe, zero API costs for sending.
Eingebaute Discord-Webhook- und Telegram-Bot-Module. Sende Rich-Embeds mit KI-Code-Reviews aus der CI, poste Benachrichtigungen oder baue eigene Chat-Bots — reines Pipe, keine API-Kosten fürs Senden.
import "discord.pipe" as d ai_provider "deepseek" -- Send AI code review to Discord on every push review: ai_chat "Review this code" diff 800 d.d_webhook_embed (env "DISCORD_WEBHOOK") { title: "🔧 CI: Push to master", color: 3447003, fields: [ {name: "Changed Files", value: stat}, {name: "AI Review", value: review} ] }
Same job. Less code. Built-in safety.
Gleicher Job. Weniger Code. Eingebaute Sicherheit.
| Python + LangChain | Pipe | |
|---|---|---|
| RAG pipelineRAG-Pipeline | 26 LOC26 Zeilen | 14 LOC14 Zeilen |
| Sandbox LLM accessLLM-Zugriff sandboxen | Custom middlewareCustom Middleware | One sandbox_profile blockEin sandbox_profile-Block |
| Switch AI providerKI-Provider wechseln | Rewrite SDK callsSDK-Calls umschreiben | ai_provider "deepseek" |
| Deploy to serverAuf Server deployen | Docker + venv + pipDocker + venv + pip | scp pipe binaryscp pipe binary |
| Expose an HTTP APIHTTP-API bereitstellen | FastAPI + uvicorn + depsFastAPI + uvicorn + Deps | route_get + serve |
| Parallel LLM callsParallele LLM-Calls | asyncio.gather() boilerplateasyncio.gather()-Boilerplate | >> operator, ai_batch |
| MCP Server + ClientMCP Server + Client | langchain-mcp-adapters + pip depslangchain-mcp-adapters + pip-Deps | 13 MCP builtins, zero deps, any stdio server |
| Binary size (with deps)Binary-Größe (mit Deps) | 345 MB venv345 MB venv | 8.6 MB8.6 MB |
36 AI operations are language primitives — not library calls. summarize, translate, classify work without imports, SDKs, or API wrappers.
36 KI-Operationen sind Sprach-Primitives — keine Library-Calls. summarize, translate, classify funktionieren ohne Imports, SDKs oder API-Wrapper.
Declarative sandbox profiles restrict exec, write_file, and http_get. Essential for ai_with_tools — keep LLMs on a leash.
Deklarative Sandbox-Profile beschränken exec, write_file und http_get. Essentiell für ai_with_tools — LLMs an die Leine nehmen.
One statically-linked ~8 MB binary — no venv, no pip, no Docker. pipe -build embeds your script + modules; -upx compresses to ~2.9 MB. Linux, macOS, Windows, Raspberry Pi — or your browser via WebAssembly.
Eine statisch gelinkte ~8-MB-Binary — kein venv, kein pip, kein Docker. pipe -build bettet Skript + Module ein; -upx komprimiert auf ~2,9 MB. Linux, macOS, Windows, Raspberry Pi — oder dein Browser per WebAssembly.
>> starts any pipeline stage in the background. Futures auto-resolve. ai_batch processes hundreds of texts concurrently with rate limiting.
>> startet jede Pipeline-Stufe im Hintergrund. Futures lösen sich automatisch auf. ai_batch verarbeitet hunderte Texte parallel mit Rate-Limiting.
OpenAI, Anthropic, DeepSeek, Ollama, OpenRouter. Switch providers with ai_provider. Same code. Same pipeline. Zero rewrites.
OpenAI, Anthropic, DeepSeek, Ollama, OpenRouter. Provider wechseln mit ai_provider. Gleicher Code. Gleiche Pipeline. Keine Rewrites.
Zero-setup testing: test blocks with assert_eq, assert_error — run via pipe -test. The official GitHub Action runs Pipe in CI with sandbox profiles on demand.
Testen ohne Setup: test-Blöcke mit assert_eq, assert_error — ausgeführt per pipe -test. Die offizielle GitHub Action führt Pipe in CI aus, Sandbox bei Bedarf.
23 curated modules: discover with pipe -search, install with pipe -get, pin versions with @1.0.0. Plus LSP-powered IntelliSense in VSCode — completion, hover docs, go-to-definition.
23 kuratierte Module: entdecken mit pipe -search, installieren mit pipe -get, Versionen pinnen mit @1.0.0. Dazu LSP-IntelliSense in VSCode — Completion, Hover-Docs, Go-to-Definition.
pipe-web: routing, JSON, middleware, and a real HTTP server. Ship APIs and dashboards as a single binary — no FastAPI, no Express. See the AI Summarize API and RAG Knowledge Base examples.
pipe-web: Routing, JSON, Middleware und ein echter HTTP-Server. APIs und Dashboards als einzelne Binary ausliefern — kein FastAPI, kein Express. Siehe AI-Zusammenfassungs-API und RAG-Wissensbasis.
A faithful Pipe core that runs on MicroPython and a real ESP32 — same pipelines, same MCP client over WiFi. Read a sensor, call an MCP server, drive a GPIO, all in Pipe. Preview, MIT: github.com/MachuraHarry/micropipe
Ein treuer Pipe-Kern, der auf MicroPython und einem echten ESP32 läuft — gleiche Pipelines, gleicher MCP-Client über WLAN. Sensor lesen, MCP-Server aufrufen, GPIO schalten — alles in Pipe. Preview, MIT: github.com/MachuraHarry/micropipe
✓ Officially listed in the official MCP Registry — one-click install from GitHub MCP Registry for Copilot & VS Code ✓ Offiziell im offiziellen MCP-Registry gelistet — One-Click-Install aus der GitHub-MCP-Registry für Copilot & VS Code
| summarize | Text summarization |
| translate | Translation |
| classify | Classification |
| extract | Data extraction (JSON) |
| ask | Question answering |
| generate | Free-text generation |
| ai_stream | Real-time token streaming |
| ai_batch | Auto-parallel batch |
| ai_parallel | Concurrency control |
| ai_rate_limit | Rate limiting |
| ai_chat | Low-level chat |
| ai_chat_json | Chat → structured JSON |
| embed | Text → vector |
| embed_batch | Batch embeddings |
| cosine_sim | Semantic similarity |
| dot_product | Dot product |
| nearest | Top-K nearest |
| ai_tool | Register function as tool |
| ai_with_tools | Chat with tool access |
| ai_provider | Select AI provider |
| ai_model | Select model |
| ai_timeout | Set timeout |
| swarm_agent | Register a swarm member |
| ai_swarm | Run a handoff multi-agent swarm |
| ai_swarm_trace | Run a swarm, with trace |
| ai_vision | Answer questions about an image |
| mcp_server | Create MCP server |
| mcp_serve_stdio | Start stdio server |
| mcp_serve_sse | Start HTTP server |
| mcp_tools | List tools |
| mcp_resource | Define resource |
| mcp_resource_template | Dynamic resource |
| mcp_prompt | Define prompt |
| mcp_resources | List resources |
| mcp_read_resource | Read resource |
| mcp_prompts | List prompts |
| mcp_prompt_get | Get prompt |
| mcp_use_stdio | Connect to MCP server |
| mcp_use_sse | Connect via HTTP |
| sandbox_profile | Define a sandbox profile |
| set_sandbox | Activate a profile |
| with_sandbox | Temp profile override |
| sandbox_lock | Lock sandbox state |
| test | Grouped test block |
| assert | Truthy check |
| assert_eq | Equality check |
| assert_lt | Less-than check |
| assert_gt | Greater-than check |
| assert_error | Expect an error |
Write and run Pipe code instantly. No install. No signup. Full syntax highlighting.
Pipe-Code sofort schreiben und ausführen. Keine Installation. Kein Login. Volles Syntax-Highlighting.
Open PlaygroundPre-built binaries for Linux, macOS & Windows — or build from source with git clone + make build.
Fertige Binaries für Linux, macOS & Windows — oder aus dem Quellcode mit git clone + make build.
curl -fsSL https://pipe-lang.com/install.sh | bashInstall Guide
Run Pipe in GitHub Actions. No installation. AI-enabled on demand. Add sandbox on demand via the flags input (e.g. -vm -q --sandbox).
Pipe in GitHub Actions ausführen. Keine Installation. KI bei Bedarf aktivierbar. Sandbox bei Bedarf über den flags-Input (z. B. -vm -q --sandbox).
Pipe can connect to any MCP server via mcp_use_stdio. GitHub, Filesystem, Postgres, Slack, Brave Search, Git, Memory, Sequential Thinking — everything discoverable from npm/uvx becomes a tool for the AI.
Pipe kann sich mit jedem MCP-Server via mcp_use_stdio verbinden. GitHub, Dateisystem, Postgres, Slack, Brave Search, Git, Memory, Sequential Thinking — alles, was per npm/uvx verfügbar ist, wird zum Tool für die KI.
✓ Pipe is listed in the official MCP Registry (v1.1.1) and on the GitHub MCP Registry for one-click install from Copilot & VS Code. ✓ Pipe ist im offiziellen MCP-Registry gelistet (v1.1.1) und in der GitHub-MCP-Registry für One-Click-Install aus Copilot & VS Code.
Read, write, list, search files — secured to allowed directories.
Issues, PRs, repos, files — full GitHub API as AI tools.
Schema inspection, read-only queries — AI explores your data.
Web content fetching, real-time search — AI stays informed.
Channels, messages, history — integrate AI into team comms.
Persistent knowledge graph — AI remembers across sessions.
Releases, deep dives, and tutorials — straight from building with Pipe.
Releases, Deep Dives und Tutorials — direkt aus der Entwicklung von Pipe.
One command turns any Git repository into a full RAG MCP server: keyword search without API keys, cited AI answers (now with OpenRouter free models), code symbol lookup in 5 languages, file outlines — persistent SQLite indexes and a locked-down sandbox included.
Ein Befehl macht aus jedem Git-Repository einen vollwertigen RAG-MCP-Server: Keyword-Suche ohne API-Key, zitierte KI-Antworten (jetzt mit OpenRouter-Free-Models), Code-Symbol-Lookup in 5 Sprachen, File-Outlines — mit persistenten SQLite-Indexen und verriegelter Sandbox.
A semantic search and RAG MCP server for the Pipe language — 7 tools, zero dependencies, published on the MCP Registry. Heading-aware chunking, hybrid search, cited answers, code symbol lookup.
Ein semantischer Suche- und RAG-MCP-Server für die Pipe-Sprache — 7 Tools, null Abhängigkeiten, veröffentlicht auf dem MCP Registry. Heading-bewusstes Chunking, Hybrid-Suche, zitierte Antworten, Code-Symbol-Lookup.
A documentation-native RAG module plus a web dashboard: heading-aware chunking, hybrid keyword + semantic search, cited answers, incremental re-indexing, and a one-command UI — all in pure Pipe, no vector database.
Ein dokumentations-natives RAG-Modul plus Web-Dashboard: heading-bewusstes Chunking, hybride Keyword- + semantische Suche, zitierte Antworten, inkrementelles Indexieren und ein Ein-Befehl-UI — alles in reinem Pipe, ohne Vektor-Datenbank.