Kortx Server

npm version License: MIT Build Status Node.js Version

Quick Start • Documentation • Examples • Contributing

Kortx is a lightweight Model Context Protocol (MCP) server that gives coding copilots access to:

• OpenAI GPT-5 models (gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-codex, gpt-5.1-2025-11-13, gpt-5.1-codex) with automatic fallback.
• Perplexity Sonar models for real-time research.
• GPT Image (gpt-image-1) for visual generation and editing.
• A default context gatherer that can ingest local file excerpts and optional connectors for Serena, MCP Knowledge Graph, and CCLSP MCP servers when those are running.

The server ships with structured logging, request rate limiting, response caching, and a hardened Docker build that runs as a non-root user. Transport is stdio-only today (HTTP is not implemented yet).

---

Highlights

• Seven consultation tools plus a batch runner covering planning, alternatives, copy improvement, debugging, expert consultation, research, and image workflows.
• File-based context enrichment out of the box, with pluggable MCP connectors ready for Serena/MCP Knowledge Graph/CCLSP when available.
• Perplexity integration (requires API key) for citation-backed answers and image search.
• Configurable OpenAI model, reasoning effort, verbosity, and retry behaviour.
• Built-in rate limiting, cache, and optional audit logging to avoid flooding upstream APIs.
• Dockerfile uses multi-stage build, npm audits, and runs as UID/GID 1001.

---

Quick Start

1. Set credentials (both keys are required by the current config):

   export OPENAI_API_KEY=sk-your-openai-key
   export PERPLEXITY_API_KEY=pplx-your-perplexity-key
   

2. Add Kortx to your MCP client. Example generic configuration:

   {
     "mcpServers": {
       "kortx-mcp": {
         "command": "npx",
         "args": ["-y", "@effatico/kortx-mcp@latest"],
         "env": {
           "OPENAI_API_KEY": "${OPENAI_API_KEY}",
           "PERPLEXITY_API_KEY": "${PERPLEXITY_API_KEY}"
         }
       }
     }
   }
   

Client-specific walkthroughs for Claude Code, VS Code Copilot, Cursor, and others are available under docs/integration.

---

Tool Overview

• think-about-plan – Structured review of plans with strengths, risks, and follow-up questions.
• suggest-alternative – Generates viable alternatives with trade-offs and constraints.
• improve-copy – Refines technical copy with tone, clarity, and accessibility guidance.
• solve-problem – Debugging assistant covering root cause analysis and remediation steps.
• consult – Expert consultation with domain-specific personas (software-architecture, security, performance, database, devops, frontend, backend, ai-ml, general).
• search-content – Perplexity-backed web/academic/SEC search with citations and optional images.
• create-visual – GPT Image based generator/editor; search mode reuses Perplexity for visual inspiration.
• batch-consult – Runs multiple tool calls in parallel and returns aggregated results.

Every consultation tool accepts an optional preferredModel. The OpenAI client falls back through gpt-5.1-2025-11-13 → gpt-5.1-codex → gpt-5 → gpt-5-mini → gpt-5-nano automatically on failures.

---

Configuration Essentials

Minimum environment variables:

• OPENAI_API_KEY – required
• PERPLEXITY_API_KEY – required (disable Perplexity integration by omitting the search-content / create-visual search mode if you do not have a key)

Common overrides (see .env.example for the full list):

# OpenAI behaviour
OPENAI_MODEL=gpt-5-mini        # gpt-5 | gpt-5-mini | gpt-5-nano | gpt-5-codex | gpt-5.1-2025-11-13 | gpt-5.1-codex
OPENAI_REASONING_EFFORT=minimal
OPENAI_VERBOSITY=low
OPENAI_MAX_TOKENS=1024

# Safety & performance
ENABLE_RESPONSE_CACHE=true
CACHE_MAX_SIZE_MB=100
ENABLE_RATE_LIMITING=true
MAX_REQUESTS_PER_HOUR=100

# Context gathering
ENABLE_SERENA=false            # flip to true when a Serena MCP server is reachable
ENABLE_MEMORY=false            # same for MCP Knowledge Graph
ENABLE_CCLSP=false             # same for cclsp
INCLUDE_FILE_CONTENT=true

Note The Serena, MCP Knowledge Graph, and CCLSP connectors are stubs that return data only when the corresponding MCP servers are running and reachable. Out of the box the gatherer uses on-disk file excerpts referenced in prompts.

Full reference: docs/configuration.md

---

Docker

Build and run locally:

docker build -t kortx-mcp .
docker run -i --rm \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e PERPLEXITY_API_KEY=$PERPLEXITY_API_KEY \
  kortx-mcp

The image:

• Uses Node.js 22 Alpine
• Performs npm audit during build
• Copies only the compiled build/ artefacts and production deps
• Runs as user nodejs (UID 1001)

Compose example (docker-compose.yml) is included for longer-lived runs with volume mounts.

---

Development

git clone https://github.com/effatico/kortx-mcp.git
cd kortx-mcp
npm install
cp .env.example .env
npm run build
npm run dev

Useful scripts:

• npm test / npm run test:coverage
• npm run lint / npm run lint:fix
• npm run format / npm run format:check
• npm run inspector – launch MCP Inspector for interactive debugging

Node.js ≥ 22.12.0 and npm ≥ 9 are required.

---

Documentation

• Getting Started
• Integration Guides
• API Reference
• Troubleshooting
• Examples

---

Contributing & Support

• Contributing Guide
• Security Policy
• GitHub Discussions
• Issue Tracker
• Email

---

License

MIT © Effati Consulting AB. See LICENSE.

Star History Chart