Kopern MCP Server provides 32 tools for building, testing, grading, and deploying AI agents directly from MCP-compatible clients (Claude Code, Cursor, Windsurf, etc.).
Agent Management: Create, update, delete, list, and interact with AI agents using custom system prompts, models, skills, and tools; deploy from 37 ready-made templates (BTP, Legal, E-commerce, etc.)
Grading & Optimization: Grade system prompts against test cases using 6 criteria types (output_match, schema_validation, tool_usage, safety_check, custom_script, llm_judge); create reusable grading suites; view detailed results and score history
AutoTune: Iteratively auto-optimize agent system prompts using improvement notes from grading runs until a target score is reached
Teams & Pipelines: Build multi-agent teams (parallel/sequential/conditional) and multi-step pipelines with configurable input mapping
Connectors: Deploy embeddable chat widgets; connect agents to Telegram, WhatsApp Business, Slack, inbound/outbound webhooks, Gmail/Outlook, and Google/Microsoft Calendar
Sessions & Memory: List and inspect conversation sessions with metrics; manage persistent agent memory (remember, recall, forget, list)
Utilities: Generate EU AI Act compliance reports (Articles 6, 12, 14, 52); view token usage and cost breakdowns per agent; export/import agent configurations as portable JSON
Enables connecting to Gmail for email integration, allowing agents to send/receive emails.
Enables connecting to Google Calendar for calendar integration, allowing agents to manage events.
Enables connecting to Slack workspace for messaging integration, allowing agents to send messages and interact.
Enables connecting a Telegram bot, allowing agents to send and receive messages through Telegram.
Enables connecting WhatsApp Business, allowing agents to send and receive messages through WhatsApp.
32 AI agent tools for Claude Code, Cursor, Windsurf, and any MCP client.
Build, test, grade, and deploy AI agents — all from your terminal or IDE.
claude mcp add kopern -- npx -y @kopern/mcp-serverThen set your API key:
export KOPERN_API_KEY=kpn_your_key_hereAdd to your .mcp.json (project) or ~/.claude/settings.json (global):
{
"mcpServers": {
"kopern": {
"command": "npx",
"args": ["-y", "@kopern/mcp-server"],
"env": {
"KOPERN_API_KEY": "kpn_your_key_here"
}
}
}
}Add to your MCP settings:
{
"mcpServers": {
"kopern": {
"command": "npx",
"args": ["-y", "@kopern/mcp-server"],
"env": {
"KOPERN_API_KEY": "kpn_your_key_here"
}
}
}
}If your client supports HTTP MCP transport directly:
{
"mcpServers": {
"kopern": {
"type": "http",
"url": "https://kopern.ai/api/mcp/server",
"headers": {
"Authorization": "Bearer kpn_your_key_here"
}
}
}
}Related MCP server: Google Analytics MCP Server
Go to kopern.ai and sign in
Navigate to Settings > API Keys
Click Generate API Key
Copy the kpn_... key
Two key types:
User-level key — access to all 30 platform tools (agent CRUD, grading, templates, etc.)
Agent-bound key — access to all 32 tools including kopern_chat and kopern_agent_info
Tool | Description |
| Create a new agent with system prompt, model, skills, and tools |
| Get full agent configuration and metadata |
| Update agent system prompt, model, domain, or builtin tools |
| Permanently delete an agent |
| List all your agents with scores and versions |
| Send a message to an agent (with tool calling) |
| Get agent metadata (agent-bound key only) |
| Deploy from 37 ready-made templates (BTP, Legal, E-commerce...) |
Tool | Description |
| Grade a system prompt against inline test cases |
| Create a reusable test suite with cases |
| Execute a grading suite, get scores + improvement notes |
| AutoTune — iterative prompt optimization (uses improvement notes) |
| Get detailed results + improvement notes for a grading run |
| List score history for a suite |
Tool | Description |
| Create multi-agent teams (parallel, sequential, conditional) |
| Execute a team with a prompt |
| Create multi-step pipelines with input mapping |
| Execute a pipeline sequentially |
Tool | Description |
| Deploy an embeddable chat widget |
| Connect a Telegram bot |
| Connect WhatsApp Business |
| Connect to Slack workspace |
| Set up inbound/outbound webhooks |
| Connect Gmail or Outlook |
| Connect Google or Microsoft Calendar |
Tool | Description |
| List conversation sessions with metrics |
| Get full session details (events, tool calls, tokens) |
| Agent memory CRUD (remember, recall, forget, list) |
Tool | Description |
| Browse 37 templates (28 general + 9 vertical) |
| Generate EU AI Act compliance report |
| Token usage, cost, per-agent breakdown |
| Export agent as portable JSON |
| Import agent from JSON export |
> Use kopern_create_agent to create a customer support agent for my SaaS product> Use kopern_grade_prompt to test my system prompt with these cases:
- "What's your refund policy?" should mention "30-day guarantee"
- "Can I cancel?" should be empathetic and provide steps> Use kopern_list_templates to show me business templates, then deploy the "restaurant" one> Run kopern_run_grading on my agent — it will return scores AND improvement notes.
Then run kopern_run_autoresearch to optimize using those suggestions automatically.> Create an agent, add a grading suite with 5 test cases, run grading,
then run autoresearch to optimize the prompt to 90%+ scoreVariable | Required | Default | Description |
| Yes | — | Your Kopern API key ( |
| No |
| Custom endpoint URL |
This package is a lightweight stdio-to-HTTP bridge. It receives JSON-RPC messages on stdin from your MCP client, forwards them to the Kopern Streamable HTTP endpoint, and returns responses on stdout.
Claude Code ──stdin──> @kopern/mcp-server ──HTTP──> kopern.ai/api/mcp/server
<──stdout── <──JSON──The Kopern server handles all MCP protocol methods (initialize, tools/list, tools/call, ping). Zero dependencies — just Node.js 18+ and fetch.
MIT
Generate an EU AI Act compliance report for an agent. Checks Art. 6 (risk), Art. 12 (audit trail), Art. 14 (human oversight), Art. 52 (transparency). No LLM cost.
Inputs: agent_id (required).
Connect an agent to Google Calendar or Microsoft Calendar for scheduling tools (list_events, create_event, etc.). Requires OAuth in browser. Enables the service_calendar builtin tool.
Inputs: agent_id (required), provider (required).
Connect an agent to Gmail or Outlook for email tools (read_emails, send_email, reply_email). Requires OAuth in browser. Enables the service_email builtin tool.
Inputs: agent_id (required), provider (required).
Connect an agent to Slack. Returns an OAuth install URL to authorize in your browser (Slack requires interactive OAuth).
Inputs: agent_id (required).
Connect an agent to Telegram via a bot. Requires a bot token from @BotFather.
Inputs: agent_id (required), bot_token (required).
Create an inbound or outbound webhook for an agent. Inbound: receive messages via HTTP POST. Outbound: send events to your URL (n8n, Zapier, Make compatible).
Inputs: agent_id (required), name, type, events, +2 more.
Connect an agent to WhatsApp Business. Requires Meta Cloud API credentials.
Inputs: agent_id (required), access_token (required), phone_number_id (required), phone_number, +1 more.
Enable the embeddable chat widget for an agent. Returns the embed code for your website.
Inputs: agent_id (required), position, allowed_origins, welcome_message.
Create a new AI agent with a system prompt, model, and optional skills. Returns the agentId.
Inputs: name (required), system_prompt (required), model, domain, +4 more.
Create a grading suite with test cases on an agent. Each case has an input prompt and expected behavior for evaluation.
Inputs: cases (required), agent_id (required), name, description.
Create a multi-step pipeline on an agent. Steps chain agents sequentially with configurable input mapping.
Inputs: name (required), steps (required), agent_id (required), description.
Create a multi-agent team. Agents work together in parallel, sequential (chain), or conditional (router) mode.
Inputs: name (required), agents (required), description, execution_mode.
Permanently delete an agent and all its data (skills, tools, grading suites, sessions, connectors).
Inputs: agent_id (required).
Deploy an agent from a template (28 general + 9 vertical). Creates agent + skills + tools + grading suite in one shot. Use kopern_list_templates to see available slugs.
Inputs: slug (required), answers.
Export an agent as a portable JSON object (agent config, skills, tools, extensions, grading suites with cases). Use kopern_import_agent to re-import. No LLM cost.
Inputs: agent_id (required).
Get full details of an agent: system prompt, model, skills count, tools count, grading suites count.
Inputs: agent_id (required).
Get detailed results of a grading run: per-case scores, agent outputs, criteria evaluations, improvement notes. No LLM cost.
Inputs: run_id (required), agent_id (required), suite_id (required).
Get full details of a session including message events, tool calls, and metrics. No LLM cost.
Inputs: agent_id (required), session_id (required).
Get token usage and cost metrics. Shows input/output tokens, cost, grading runs, and per-agent breakdown. No LLM cost.
Inputs: year_month, include_history.
Grade a system prompt against inline test cases. Uses 6 criteria types (output_match, schema_validation, tool_usage, safety_check, custom_script, llm_judge). Returns score 0-1. Uses YOUR API keys.
Inputs: test_cases (required), system_prompt (required), model, provider.
Import an agent from a Kopern export JSON. Creates a new agent with all skills, tools, extensions, and grading suites. No LLM cost.
Inputs: data (required).
List all your Kopern agents (name, description, model, domain, grading score). No LLM cost.
Inputs: no named parameters.
List grading runs for a suite. Shows score history, pass rates, and versions over time. No LLM cost.
Inputs: agent_id (required), suite_id (required).
List conversation sessions for an agent. Shows purpose, source, token usage, cost, timestamps. No LLM cost.
Inputs: agent_id (required), limit.
List all 37 AI agent templates (28 general + 9 vertical/business). Returns slug, title, domain, tagline. No LLM cost.
Inputs: category.
Manage an agent's persistent memory. Actions: remember (save key-value), recall (search by query), forget (delete by key), list (all memories). No LLM cost.
Inputs: action (required), agent_id (required), key, query, +2 more.
Run AutoTune optimization on an agent. Iteratively mutates the system prompt, re-grades, and keeps improvements. Returns the optimized score. Uses YOUR API keys. Can take several minutes.
Inputs: agent_id (required), suite_id (required), target_score, max_iterations, +1 more.
Run a grading suite on an agent. Executes all test cases, evaluates with configured criteria, returns detailed scores. Uses YOUR API keys.
Inputs: agent_id (required), suite_id (required).
Execute a pipeline on a prompt. Steps run sequentially, each feeding its output to the next. Uses YOUR API keys.
Inputs: prompt (required), agent_id (required), pipeline_id (required).
Execute a multi-agent team on a prompt. Returns each agent's output and the final combined result. Uses YOUR API keys.
Inputs: prompt (required), team_id (required).
Update any part of an agent: config, skills, tools, or extensions. Use add/remove arrays for granular control over subcollections.
Inputs: agent_id (required), name, model, domain, +10 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 31 tools
Each tool has a clearly distinct purpose, from agent lifecycle management to external service connections and evaluation. Even similar tools like kopern_create_agent and kopern_deploy_template are differentiated by the template-based bulk creation. No significant overlap.
All tools follow the consistent pattern 'kopern_verb_noun' in lowercase snake_case. Verbs like connect, create, delete, get, list, run, update are used predictably, making the toolset easy to navigate.
With 31 tools, the server is on the higher end but justifiably so due to the broad scope covering agent creation, connections, grading, pipelines, teams, and more. The count is appropriate for the comprehensive feature set without feeling bloated.
The tool surface appears complete for an agent management server: full CRUD for agents, integration with multiple external services, grading, optimization, memory management, and team/pipeline execution. No obvious missing operations for the stated domain.
Build, validate, and deploy multi-agent AI solutions from any AI environment.
Train, evaluate, red-team and deploy AI agents: prompt tuning, LLM-judged evals, katas.
The AI orchestration agent for modern software teams.
Build, deploy, and sell AI agents for local-service businesses - from your IDE.
