Skip to main content
Glama
kondukt-dev

Kondukt

Official
by kondukt-dev

Kondukt

Postman for MCP servers — test, validate, debug, and scaffold.

npm version npm downloads License: MIT MCP Compatible


Kondukt is an open-source developer toolkit for the Model Context Protocol (MCP). It connects to any MCP server, inspects what it exposes, validates protocol compliance, calls tools interactively, and scaffolds new servers from templates.

And it's itself an MCP server — so you can register it with Claude Code, Codex, or Gemini CLI, and ask your AI agent to test and validate other MCP servers for you.

Kondukt demo

Status: v0.1.x. Stable enough for daily use, but still evolving. Bug reports and PRs welcome.

Why Kondukt

MCP has 97M+ SDK downloads and 10,000+ published servers. Every major AI vendor ships it. And yet, building an MCP server today feels like writing HTTP APIs in 2005 — no Postman, no linter, no scaffolder. The only existing tool (MCP Inspector) is a basic debugger; there's no validation, no scoring, no project generation.

Kondukt is the tool I wanted and couldn't find. If you build MCP servers, it should save you hours.

MCP Inspector

Kondukt

Inspect tools/resources/prompts

Call tools interactively

Protocol validation

✅ (18 rules)

Quality score

✅ (0–100)

Scaffold new servers

Usable from AI agents

✅ (itself an MCP server)

AI context file generation

Two ways to use it

As a CLI — fast feedback while you build:

npx kondukt test "npx -y @modelcontextprotocol/server-everything"
npx kondukt validate "npx -y @modelcontextprotocol/server-everything"

As an MCP server — your agent does the work:

claude mcp add kondukt -- npx kondukt serve

Then, inside Claude Code:

"Test the MCP server at npx my-server and tell me what tools it exposes. Then run a validation and fix anything that fails."

Claude calls Kondukt's tools directly. No context-switching between terminal and editor.

Quick start

No install needed — everything runs via npx.

# Discover tools, resources, and prompts on any MCP server
npx kondukt test "npx -y @modelcontextprotocol/server-everything"

# Run 18 protocol compliance checks and get a 0–100 score
npx kondukt validate "npx -y @modelcontextprotocol/server-everything"

# Call a specific tool with arguments
npx kondukt call "npx -y @modelcontextprotocol/server-everything" \
  --tool echo --args '{"message": "hello"}'

# Generate a new MCP server project
npx kondukt scaffold my-server --template typescript \
  --tool "get_weather:Get weather for city:city:string"

# Generate AI context files (CLAUDE.md, AGENTS.md, GEMINI.md) for this repo
npx kondukt agent-docs . --all

Features

kondukt test — inspect any MCP server

Connects to a server (stdio or HTTP/SSE), auto-discovers tools, resources, and prompts, and prints them in a structured, readable format.

kondukt validate — 18 protocol rules, 0–100 score

Four categories:

  • Tools — schema validity, descriptions, duplicate names

  • Resources — URI formats, MIME types, metadata

  • Prompts — argument schemas, descriptions

  • Protocol — initialization, capabilities, error handling

Every violation comes with a rule ID, severity, and a suggested fix.

kondukt validate

kondukt call — interactive debugging

Execute any tool exposed by a server with custom arguments. See the raw response. Essential for reproducing bugs.

kondukt call

kondukt scaffold — new server in 10 seconds

Generates a complete, runnable MCP server project. TypeScript or Python templates. Includes types, tests, CI config, and README. Define tools from the CLI:

npx kondukt scaffold weather-server \
  --template typescript \
  --tool "get_forecast:Get forecast:city:string,days:number"

kondukt scaffold

kondukt agent-docs — context files for AI coding tools

Analyzes a codebase and generates CLAUDE.md, AGENTS.md, or GEMINI.md. Detects frameworks, ORMs, test runners, and project conventions via static analysis.

kondukt agent-docs

kondukt serve — run Kondukt as an MCP server

Exposes all of the above as MCP tools. Register once, use from any MCP-compatible agent.

Install

Most commands work with zero install via npx. For frequent use:

npm install -g kondukt

Or use it as a library:

npm install kondukt
import { McpConnection, SchemaValidator } from "kondukt";

const conn = new McpConnection({
  type: "stdio",
  command: "npx",
  args: ["-y", "@modelcontextprotocol/server-everything"],
});

await conn.connect();
const result = await new SchemaValidator().validate(conn);
await conn.disconnect();

console.log(result.score); // 0–100
console.log(result.issues);

Transports

Kondukt supports both transports defined by the MCP spec:

  • stdio — pass the command directly: npx kondukt test "npx -y my-server"

  • HTTP / SSE — pass a URL: npx kondukt test "https://my-server.example.com/mcp"

Development

Regenerating the demo assets

./scripts/record-demo.sh                     # all scenarios
./scripts/record-demo.sh demo scaffold       # a subset
./scripts/record-demo.sh --upload demo       # also upload to asciinema.org

Requires Homebrew (installs asciinema, agg, and ffmpeg on demand). Each scenario in scripts/scenarios/<name>.sh produces three outputs:

  • assets/<name>.gif — committed, embedded in this README

  • assets/local/<name>.cast — asciinema source, gitignored

  • assets/local/<name>.mp4 — H.264 fallback for Twitter/X, gitignored

Only the GIFs ship in the repo; regenerating the cast and MP4 is a local concern.

Contributing

Issues, PRs, and feedback all welcome.

If you find a bug, the fastest path is an issue with a reproducing command. If Kondukt's validator gets something wrong, that's also a bug — file it.

License

MIT

Author

Built by Alex BurovGitHub · X/Twitter

A
license - permissive license
-
quality - not tested
B
maintenance

Maintenance

Maintainers
Response time
3dRelease cycle
2Releases (12mo)

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

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/kondukt-dev/core'

If you have feedback or need assistance with the MCP directory API, please join our Discord server