NLBackend
Provides email sending capabilities through Resend's API, allowing actions such as sending emails with subject and body.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@NLBackendcreate a product schema with name, price, stock"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
NLBackend
Define backends in natural language. Run them via MCP.
NLBackend is a framework where you describe your data models, business rules, actions, and workflows in plain Markdown files — and the framework turns them into a fully functional API that LLMs can interact with through the Model Context Protocol.
No code. Just natural language.
schema/user.md → users_create, users_get, users_list, users_update, users_delete
schema/recipe.md → recipes_create, recipes_get, recipes_list, ...
rules/permissions.md → enforced on every operation
workflows/publish.md → run_workflow("publish")How it works
┌─────────────────────────────────────────────────────────────┐
│ Your Project (Markdown) │
│ schema/*.md actions/*.md rules/*.md workflows/*.md │
└──────────────────────────┬──────────────────────────────────┘
│
┌──────▼──────┐
│ NLBackend │ ← compiles schemas, registers tools
│ MCP Server │ ← file-based DB, auto CRUD
└──────┬──────┘
│ stdio (MCP protocol)
┌──────▼──────┐
│ Claude / │ ← calls users_create, query_db, etc.
│ Any LLM │
└─────────────┘Two LLM roles:
Building LLM — reads
claude.mdfiles, writes.mddefinitions. Builds the backend.Consuming LLM — connects via MCP, calls tools, reads/writes data. Uses the backend.
Related MCP server: LLMling
Quick start
Prerequisites
Bun v1.0+:
curl -fsSL https://bun.sh/install | bashInstall & create a project
git clone https://github.com/your-org/nlbackend.git
cd nlbackend
bun install
# Scaffold a new project
bun run src/cli.ts init my-appDefine your data model
Create my-app/schema/task.md:
# Task
A task in a to-do list.
## Fields
- **id**: string, auto uuid, immutable
- **title**: string, required, min 1, max 200
- **done**: boolean, default false
- **created_at**: string, auto timestamp, immutable
- **updated_at**: string, auto timestampThat's it. The framework auto-generates tasks_create, tasks_get, tasks_list, tasks_update, and tasks_delete tools.
Start the server
bun run src/index.ts ./my-appConnect an LLM
bun run src/cli.ts config ./my-appThis outputs the MCP config JSON. Paste it into your client:
Claude Desktop (~/.claude/claude_desktop_config.json):
{
"mcpServers": {
"my-app": {
"command": "bun",
"args": ["run", "/path/to/nlbackend/src/index.ts", "/path/to/my-app"]
}
}
}Cursor (.cursor/mcp.json): same format.
The LLM can now call tasks_create, tasks_list, query_db, and all other tools.
Project structure
my-app/
├── project.md # Name & description
├── claude.md # Instructions for the Building LLM
├── schema/ # Data models (one .md per entity)
│ └── claude.md # Conventions for writing schemas
├── actions/ # Custom MCP tools beyond CRUD
│ └── claude.md
├── rules/ # Business rules & validation
│ └── claude.md
├── workflows/ # Multi-step processes (saga pattern)
│ └── claude.md
├── integrations/ # External service configs (email, webhooks)
│ └── claude.md
├── config/server.md # LLM provider settings
├── tests/ # Natural language test scenarios
│ └── claude.md
└── db/ # Auto-managed file databaseEvery folder has a claude.md that teaches an LLM how to write files for that folder. Share the project with Claude and describe what you want — it knows the conventions.
What you get automatically
You write | Framework provides |
|
|
| Same 5 CRUD tools for recipes |
| Custom |
| Enforced business rules |
|
|
Nothing |
|
Schema keywords
Schemas are compiled with a rule-based parser (no LLM needed). Use these recognized keywords:
Keyword | Example |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Cannot be changed after creation |
System tools
These are always available on every NLBackend server:
Tool | Purpose |
| Returns all schemas, tools, and compilation status |
| Read with filters, sorting, pagination |
| Low-level create/update/delete |
| View compiled state of any schema, action, rule, or workflow |
| Trigger LLM compilation of actions/rules/workflows |
| Dry-run — shows what would happen without executing |
| Execute a multi-step workflow |
MCP resources
The server exposes read-only resources for the consuming LLM:
Resource URI | Content |
| Full project overview, data model, available tools, getting-started guide |
| Detailed schema for a specific entity |
CLI
nlbackend <project-path> # Start the MCP server
nlbackend <project-path> --compile # Start with LLM compilation
nlbackend init [<folder>] # Create a new project from template
nlbackend config [<project-path>] # Output MCP client connection config
nlbackend test [<project-path>] # Run .test.md natural language tests
nlbackend version # Print versionAdvanced features
Actions (custom tools)
For operations beyond CRUD, create action files in actions/{entity}/:
# Search Recipes
> Tier: 2
> Auth: public
Searches recipes by keyword, cuisine, or ingredients.
## Input
- **query**: string, optional — keyword search
- **cuisine**: string, optional — filter by cuisine type
- **max_time**: integer, optional — max cooking time in minutes
## Output
Returns matching recipes sorted by relevance.Actions are LLM-compiled into execution plans. Requires ANTHROPIC_API_KEY and --compile flag.
Rules
Define business rules in rules/*.md:
# Permissions
## Only owners can edit
A user can only update or delete a recipe if they are the author.
## Admin override
Users with role "admin" can update or delete any record.Workflows
Multi-step processes with saga-pattern compensation:
# Publish Recipe
## Trigger
When a recipe's status changes to "published".
## Steps
1. Validate all required fields are present
2. Generate a URL-friendly slug from the title
3. Send notification email to followers
4. Update recipe status to "published"
## On failure
If any step fails, revert the status to "draft".Integrations
Connect external services in integrations/*.md:
# Email Integration
## Provider
Resend (https://api.resend.com)
## Authentication
API key stored in environment variable RESEND_API_KEY
## Available Actions
### Send Email
- **to**: email address (required)
- **subject**: text (required)
- **body**: text or html (required)Natural language tests
Write tests in .test.md with Given/When/Then:
# User CRUD Tests
## Create a user
- Given an authenticated user with role "admin"
- When calling users_create with:
- username: "alice"
- email: "alice@example.com"
- Then response contains field "id"
- And response field "username" equals "alice"Run with: nlbackend test ./my-app
Example project
See example-recipes/ for a complete Recipe Sharing Platform with users, recipes, reviews, favorites, search actions, and an email integration.
Development
bun install
bun test # 62 unit tests
bun run typecheck # TypeScript strict mode
# Run the example project
bun run src/index.ts ./example-recipes
# Run the smoke test (29 end-to-end tests)
bun run example-recipes/tests/smoke-test.tsArchitecture
Compiler — Rule-based for schemas, LLM-powered for actions/rules/workflows
Database — File-based JSON with WAL, in-memory indexes, per-collection locks
Runtime — Action executor, rule engine, workflow executor (saga pattern)
Server — MCP over stdio with auto-generated CRUD tools + system tools + resources
Cache —
.compiled/directory for warm starts without LLM calls
License
MIT
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/gfendres/NLBackend'
If you have feedback or need assistance with the MCP directory API, please join our Discord server