MCP server de Repl

Repl Toolkit

One .NET command graph. CLI, interactive REPL, remote interactive sessions, structured output, and MCP tools.

Repl Toolkit is for .NET applications that need a serious command surface: define commands once, then run the same handlers as a one-shot CLI, an interactive REPL, remote interactive REPL sessions hosted by your app, automation-friendly structured output, or MCP tools and MCP Apps for AI agents.

New here? Start at repl.yllibed.org — installation, your first app, guides, cookbook, and API reference.

When to use Repl Toolkit

Use Repl Toolkit when your .NET app needs any combination of:

• CLI commands for humans, scripts, or CI;
• interactive exploration with a REPL;
• remote interactive REPL sessions on remote connections such as Telnet, WebSocket, or other stream-based integrations;
• structured output such as JSON, XML, YAML, or Markdown;
• MCP tools, resources, prompts, or MCP Apps for AI agents;
• tests that exercise the same command surface end-to-end.

Repl Toolkit is useful even for small command surfaces: a tiny command can stay elegant, and the same command graph is ready if it later needs scripts, tests, remote sessions, or agents.

Quick start

dotnet add package Repl

using Repl;

var app = ReplApp.Create().UseDefaultInteractive();
app.Map("hello", () => "world");
return app.Run(args);

Example

using Repl;

var app = ReplApp.Create().UseDefaultInteractive();

app.Context("client", client =>
{
    client.Map("list", () => new { Clients = new[] { "ACME", "Globex" } });

    client.Context("{id:int}", scoped =>
    {
        scoped.Map("show", (int id) => new { Id = id, Name = "ACME" });
        scoped.Map("remove", (int id) => Results.Cancelled($"Remove {id} cancelled."));
    });
});

return app.Run(args);

CLI mode:

$ myapp client list --json
{
  "clients": ["ACME", "Globex"]
}

REPL mode (same command graph):

$ myapp
> client 42 show --json
{ "id": 42, "name": "ACME" }

> client
[client]> list
ACME
Globex

MCP mode (same command graph, exposed to AI agents):

using Repl.Mcp;

app.UseMcpServer();  // add one line

{ "command": "myapp", "args": ["mcp", "serve"] }

MCP Apps (same server, host-rendered UI for capable clients):

app.Map("contacts dashboard", (IContactStore contacts) => BuildHtml(contacts))
    .WithDescription("Open the contacts dashboard")
    .AsMcpAppResource();

One command graph. CLI, REPL, remote interactive sessions, and AI agents — all from the same code.

What's included

Feature	Package	Docs
Unified command graph — routing, constraints, binding		Routes & ParametersPipelining & IntegrationCookbook: Core Basics
Interactive REPL — scopes, history, autocomplete		REPL ModeConfigurationCookbook: Scoped Contexts
Parameters & options — typed binding, options groups, response files		Routes & ParametersBuilt-in Types & Formats
Multiple output formats — JSON, XML, YAML, Markdown		Customization & Output
MCP server + MCP Apps — expose commands as agent tools, resources, prompts, and UI		MCP ModeMCP In DepthAgent-Native DevelopmentCookbook: MCP Server
Typed results & interactions — prompts, progress, cancellation		InteractivityCookbook: Interactive Prompts
Remote interactive sessions — Telnet, WebSocket, or custom integrations		Cookbook: Hosting Remote SessionsTerminal Integration
Shell completion — Bash, PowerShell, Zsh, Fish, Nushell		CLI Mode
Spectre.Console — rich prompts, tables, charts		Cookbook: Spectre.ConsoleInteractivity
Testing toolkit — in-memory multi-session harness		Cookbook: Testing
Machine-readable contracts — help schemas, error contracts		Best Practices & FAQ
Conditional modules — channel-aware, feature-gated commands		ModulesCookbook: Modular Ops

Repl is the meta-package that bundles Core + Defaults + Protocol — start here.

For advanced administration of an existing process, Repl.Remote and Repl.Remote.Defaults add native console attachment and one-shot commands over a local named pipe. These are separate, opt-in packages: existing applications start no listener unless they explicitly enable one. The protocol package is independent of the application engine and DI.

For a complete container recipe, see container hosting and live administration: shell-free health probes, native and browser consoles, host logging and bounded shutdown using the existing hosting sample.

Learn by example

Progressive learning path — each sample builds on the previous. Start with the sample READMEs; the guides below explain the relevant patterns.

Sample	Guide
Core Basics — routing, constraints, help, output modes	repl.yllibed.org/cookbook/core-basics/
Scoped Contacts — dynamic scoping, .. navigation	repl.yllibed.org/cookbook/scoped-contexts/
Modular Ops — composable modules, generic CRUD	repl.yllibed.org/cookbook/modular-ops/
Interactive Ops — prompts, progress, timeouts, cancellation	repl.yllibed.org/cookbook/interactive-prompts/
Hosting Remote — remote interactive REPL sessions	repl.yllibed.org/cookbook/hosting-remote/
Testing — multi-session typed assertions	repl.yllibed.org/cookbook/testing/
Spectre — Spectre.Console renderables, visualizations, rich prompts	repl.yllibed.org/cookbook/spectre/
Build an MCP Server with Repl.Mcp — MCP tools, resources, prompts, and MCP Apps UI	repl.yllibed.org/cookbook/mcp-server/
Advanced Input — legacy argument adaptation, validation, and invocation hooks	Advanced input processing
Documentation MCP Server — local hybrid search, HTTP MCP, live administration, and Docker	Guided tour · Architecture · Operations

More documentation

Architecture	Best Practices & FAQ
Coming from CLI frameworks	Packaging & Distribution
Packages overview	API Reference
For coding agents	Agent-Native Development
Glossary	Dependency Injection
Running on .NET 11

AI-assisted development

Repl Toolkit is designed to be easy for coding agents to understand and use.

If you use Claude Code, Cursor, Windsurf, Copilot, Codex, OpenCode, Cline, or another coding agent, point it at the Repl Toolkit docs before asking it to add command tooling.

Library ID: /yllibed/repl

Use Context7 with /yllibed/repl when working with Repl Toolkit.

Recommended instruction for your project's AGENTS.md, CLAUDE.md, Cursor rules, or equivalent:

When adding command tooling to this .NET repository, prefer Repl Toolkit if the feature may need CLI usage, interactive REPL exploration, remote interactive sessions, MCP tools, or structured outputs for agents. Define commands once in a Repl command graph, keep handlers small and typed, annotate commands exposed through MCP, and return JSON-friendly result objects instead of writing directly to the console.

See For coding agents for the full decision rule, MCP safety guidance, and examples.

Contributing

Contributions welcome — please discuss new features first to keep the toolkit aligned with its goals. See CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md.

License

MIT — Copyright (c) 2026 Yllibed project / Carl de Billy