Latitude MCP Server

AI-powered prompt management for Latitude.so via Model Context Protocol

Manage PromptL prompts directly from Claude, Windsurf, or any MCP client. Features intelligent validation, dynamic tool descriptions, and git-style versioning.

✨ Key Features

• 🤖 Smart Validation - Client-side PromptL validation with AST-powered error messages

• 📋 Dynamic Descriptions - Tools show available prompts and their parameters automatically

• 🔄 Full Sync - Push/pull with automatic conflict resolution

• 🎯 Atomic Operations - Validate ALL before pushing ANY (all-or-nothing)

• 📚 52 Doc Topics - Comprehensive PromptL syntax guide with semantic search

• 🏷️ Git-Style Versioning - Name your changes like commits (feat/add-auth, fix/typo)

• ⚡ Zero Config - Just set LATITUDE_API_KEY and go

Quick Start

Installation

Configuration

Set environment variables:

Get your API key from Latitude Settings.

Usage with MCP Client

Add to your MCP client config (e.g., Claude Desktop):

🛠️ Available Tools (7)

🎯 What Makes This Special?

Dynamic Tool Descriptions - The MCP server updates tool descriptions in real-time:

• run_prompt shows: "my-prompt" (params: name, email, company)

• add_prompt shows: "Available prompts (10): prompt-a, prompt-b, ..."

Your AI assistant sees exactly what prompts exist and what parameters they need!

🚀 Real-World Workflows

Workflow 1: New Project Setup

What you see:

Workflow 2: Add New Prompt (with Dynamic Guidance)

Dynamic Description Shows:

Result:

Workflow 3: Run Prompt (with Parameter Discovery)

Dynamic Description Shows:

Result:

Workflow 4: Validation Catches Errors

Validation Error (Before ANY API Call):

2: model: gpt-4 3: --- 4: Nested!

Workflow 5: Full Sync (Initialization)

Result:

📚 Documentation Topics (52)

Core Syntax (12)

overview, structure, variables, conditionals, loops, references, tools, chains, agents, techniques, agent-patterns, mocking

Configuration (8)

config-basics, config-generation, config-json-output, config-advanced, providers-openai, providers-anthropic, providers-google, providers-azure

Messages (2)

messages-roles, messages-multimodal

Tools (4)

tools-builtin, tools-custom, tools-schema, tools-orchestration

Techniques (12)

technique-role, technique-few-shot, technique-cot, technique-tot, technique-react, technique-self-consistency, technique-constitutional, technique-socratic, technique-meta, technique-iterative, technique-step-back, technique-rag

Recipes (8)

recipe-classification, recipe-extraction, recipe-generation, recipe-chatbot, recipe-rag, recipe-analysis, recipe-moderation, recipe-support

Guides (6)

conversation-history, guide-debugging, guide-safety, guide-performance, guide-testing, guide-versioning

🛠️ Development

Build

Testing with MCP Inspector

Local Development

Project Structure

Environment Variables

PromptL Syntax Overview

PromptL is a templating language for AI prompts:

Key Features:

• YAML config header (provider, model, temperature)

• Message tags (<system>, <user>, <assistant>)

• Variables ({{ name }})

• Conditionals ({{ if }}, {{ else }})

• Loops ({{ for item in items }})

• Tools (function calling)

• Chains (multi-step <step>)

• Agents (autonomous type: agent)

Use docs({ action: "get", topic: "overview" }) for complete guide.

📖 Tool Reference

list_prompts()

List all prompts in LIVE version.

Returns: Array of prompt names with project ID

Example:

get_prompt({ name })

Get full prompt content by name.

Parameters:

• name (string) - Prompt name

Returns: Full PromptL content with config and messages

Example:

run_prompt({ name, parameters })

🎯 Dynamic: Execute a prompt. Tool description shows all prompts with their parameters!

Parameters:

• name (string) - Prompt name

• parameters (object, optional) - Input parameters

Returns: AI response with token usage

Dynamic Description:

Example:

add_prompt({ prompts?, filePaths?, versionName? })

🎯 Dynamic: Add or update prompts without deleting others. Tool description shows available prompts!

Behavior: If prompt exists → overwrites. If new → adds. Never deletes other prompts.

Parameters (choose one):

Option A - Direct content:

• prompts (array) - Array of { name, content }

Option B - From files:

• filePaths (array) - Array of paths to .promptl files

Common:

• versionName (string, optional) - Git-style name like feat/add-auth or fix/typo

Returns: Summary of added/updated prompts

Example:

push_prompts({ prompts?, filePaths?, versionName? })

🔄 FULL SYNC: Replace ALL prompts in LIVE. Deletes remote prompts not in your list.

Use for: Initial setup, complete sync, resetting LIVE to match local.

Parameters (choose one):

Option A - Direct content:

• prompts (array) - Array of { name, content }

Option B - From files:

• filePaths (array) - Array of paths to .promptl files

Common:

• versionName (string, optional) - Git-style name like feat/initial-setup

Returns: Summary of added/modified/deleted prompts

Example:

pull_prompts({ outputDir? })

🔄 FULL SYNC: Download all prompts from LIVE. Deletes existing local .promptl files first.

Use for: Initial clone, resetting local to match LIVE.

Parameters:

• outputDir (string, optional) - Output directory (default: ./prompts)

Returns: List of downloaded files

Example:

docs({ action, topic?, query? })

Access comprehensive PromptL documentation (52 topics).

Parameters:

• action (string) - "help" (overview), "get" (topic), or "find" (search)

• topic (string, optional) - Topic name for "get"

• query (string, optional) - Search query for "find"

Returns: Documentation content

Examples:

✅ Validation Features

Client-Side Validation with AST

All write operations (add_prompt, push_prompts) validate prompts before making API calls using the official promptl-ai library.

Benefits:

• ⚡ Fast feedback - No wasted API calls

• 🎯 Precise errors - Exact line and column numbers

• 📝 Code frames - See surrounding context with ^~~~ pointer

• 🤖 LLM-actionable - Errors include root cause and fix suggestions

Atomic Operations

Validate ALL, push ALL or NOTHING:

Error Message Example

105: ## EXAMPLES 106: 107:

108: questions: 109: - id: q1

Supported Error Types

• message-tag-inside-message - Nested role tags

• content-tag-inside-content - Nested content tags

• config-not-found - Missing YAML frontmatter

• invalid-config - Malformed YAML

• unclosed-block - Missing closing tag

• variable-not-defined - Undefined variable

• invalid-tool-call-placement - Tool call outside <assistant>

• ...and more from official promptl-ai compiler

🔄 Migration Guide (v2 → v3)

Tool Changes

Migration:

🔧 Troubleshooting

"Validation Failed" Errors

Problem: Prompt fails with nested tag error

Solution: The error shows exact location with code frame:

Follow the fix suggestion - errors are LLM-actionable!

"No Changes Made" After Push

Problem: push_prompts reports no changes

Cause: All prompts are already up to date (content matches LIVE)

Solution: This is normal - no action needed

Version Naming Best Practices

Good:

• feat/add-sentiment-analyzer

• fix/typo-in-greeting

• refactor/simplify-prompts

• docs/update-examples

Avoid:

• test (too vague)

• update (what was updated?)

• v1.2.3 (use semantic versioning elsewhere)

Dynamic Descriptions Not Updating

Problem: Tool descriptions show old prompt list

Cause: Cache not refreshed (30s TTL)

Solution: Wait 30 seconds or restart MCP server

Contributing

1. Fork the repository

2. Create a feature branch

3. Make your changes

4. Run npm run build to verify

5. Submit a pull request

License

ISC License - see LICENSE file for details

Links

• Latitude Platform

• Latitude Documentation

• Model Context Protocol

• npm Package

• GitHub Repository

Support

• Issues: GitHub Issues

• Documentation: Use docs({ action: "help" }) tool

• Latitude Support: Latitude Discord

Built with ❤️ for the MCP ecosystem