okareo-mcp 0.0.43

Okareo MCP server for AI model evaluation

Okareo MCP Server

The Okareo MCP server exposes Okareo's evaluation capabilities as MCP tools, allowing AI coding assistants to create and manage scenarios, register models, run evaluations, and execute multi-turn simulations directly from your editor.

For detailed documentation, see the Okareo MCP docs.

Prerequisites

• An Okareo account at app.okareo.com
• A copilot that supports MCP servers (Claude Code, Cursor, or VS Code)
• Python 3.10–3.12 (only for the local install modes; not needed for remote)

---

Remote MCP (hosted) — recommended

The fastest way to get started. No local install, no Python, no API key in .mcp.json. Browser sign-in handles auth on first connect.

Recommended (OAuth — Claude Code, Claude Desktop, Cursor, VS Code 1.101+)

Add to your copilot's MCP config (typically .mcp.json):

{
  "mcpServers": {
    "okareo": {
      "type": "http",
      "url": "https://tools.okareo.com/mcp"
    }
  }
}

Reload the copilot. It will open a browser to Okareo sign-in once; thereafter the copilot stores the token itself. Your .mcp.json contains no secrets.

Fallback (Bearer header — older clients or headless / CI)

For clients that haven't shipped the MCP OAuth flow yet, paste your API key as a bearer header. Prefer the env-var form over an inline literal:

{
  "mcpServers": {
    "okareo": {
      "type": "http",
      "url": "https://tools.okareo.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:OKAREO_API_KEY}"
      }
    }
  }
}

Working across multiple Okareo organizations

If you have access to more than one Okareo account, simply Clear Authentication and then re-authenticate with the account you want to work in. You can always ask which account your are in and have acccess to.

---

Local install (alternative)

Run the MCP server on your own machine. Useful for offline / airgapped environments and for development.

Prefer the Remote MCP section above unless you have a specific reason to install locally (airgapped, custom build, development on this repo). The remote endpoint requires no install and stays current automatically.

For multi-org users, the remote endpoint also exposes tenant management tools (list_tenants, switch_tenant).

Step 1: Set Your API Key

export OKAREO_API_KEY="your-api-key"

Add this to your ~/.zshrc or ~/.bash_profile for persistence.

Step 2: Configure Your Copilot

Claude Code

Add to .mcp.json:

{
  "mcpServers": {
    "okareo": {
      "command": "uvx",
      "args": ["okareo-mcp"],
      "env": {
        "OKAREO_API_KEY": "${OKAREO_API_KEY}"
      }
    }
  }
}

No pre-install needed — uvx handles it automatically.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "okareo": {
      "command": "uvx",
      "args": ["okareo-mcp"]
    }
  }
}

Cursor inherits OKAREO_API_KEY from your shell environment.

Alternative: pip install

If you don't have uv installed:

pip install okareo-mcp

Then use "command": "okareo-mcp" instead of "command": "uvx" with "args": ["okareo-mcp"].

---

Configuration Reference

Variable	Default	Description
OKAREO_API_KEY	(required)	Your Okareo API key
OKAREO_BASE_URL	https://api.okareo.com	Override for on-prem Okareo backend
TRANSPORT	stdio	Transport: stdio (local) or sse (Docker)
PORT	8000	Port for SSE transport

---

Available Tools

Scenarios

Tool	Description
save_scenario	Save a named scenario from rows of input/result data (idempotent)
list_scenarios	List all scenarios in the project with names, IDs, and row counts
get_scenario	Retrieve a scenario's metadata and all data rows by name or ID
create_scenario_version	Create a new version of an existing scenario with updated data
preview_delete_scenario	Preview what will be deleted before removing a scenario
delete_scenario	Permanently delete a scenario and all related test data

Generation Models

Tool	Description
list_available_llms	Browse available LLMs from the Okareo registry
register_generation_model	Register a generation model for testing by selecting an LLM from the registry
list_generation_models	List all registered generation models in the project
get_generation_model	Read detailed information about a registered generation model
update_generation_model	Change the LLM a registered generation model points to
delete_generation_model	Remove a registered generation model and all its related test data

Tests & Checks

Tool	Description
list_checks	List available quality checks (built-in and custom) for evaluating model outputs
run_test	Run a quality test that evaluates a model against a scenario using specified checks
list_test_runs	List past test runs with optional filters (model, scenario, simulation-only)
get_test_run_results	Load detailed per-row results of a test run or simulation by ID or name
get_conversation_transcript	Retrieve the full conversation transcript for a single data point
reevaluate_test_run	Re-score a completed test run against a (possibly different) set of checks
create_or_update_check	Create or update a quality check by name — model-based, code-based, or audio (upsert)
generate_check	Generate a check from a natural-language description, then save it
get_check	Retrieve a check's full configuration, including its prompt template or code
delete_check	Permanently delete a check by name

Simulations (Multi-Turn)

Tool	Description
create_or_update_target	Create or update a Target — generation model, custom endpoint, or voice (OpenAI, Deepgram, Twilio)
get_target	Retrieve a Target's configuration by name (all types)
list_targets	List all simulation targets (voice and custom_endpoint) in the project
delete_target	Remove a simulation target and all its related test data
create_or_update_driver	Define a simulated user persona that will interact with your target
get_driver	Retrieve a Driver's full configuration including the persona prompt
list_drivers	List all Driver personas in the project
list_driver_voices	Discover the voices, voice profiles, and languages available for voice drivers
run_simulation	Run a multi-turn conversation evaluation (or rerun a previous one with overrides)
list_simulations	List past simulation runs with optional filters (target, scenario, limit)

Voice Monitoring

Tool	Description
ingest_conversations	Submit completed voice conversations to Okareo for monitoring
connect_voice_integration	Connect a voice provider so its traffic flows into Okareo monitoring
list_voice_integrations	List the voice provider integrations in your project
get_voice_integration	Retrieve a voice provider integration by ID, including its status
update_voice_integration	Update a voice provider integration's metadata
rotate_voice_integration_secret	Rotate a voice provider integration's secrets
delete_voice_integration	Delete a voice provider integration by ID
get_voice_webhook_url	Get the inbound webhook endpoint for a voice provider

Analytics & Dashboards

Tool	Description
query_analytics	Query Okareo's product analytics to understand evaluation trends
list_dashboards	List the analytics dashboards in your project
get_dashboard	Retrieve a dashboard's full configuration by name
save_dashboard	Create or update an analytics dashboard by name (upsert)
reorder_dashboards	Set the display order of dashboards
delete_dashboard	Delete a dashboard by name

Tenant Management (remote MCP only)

Tool	Description
list_tenants	List every Okareo organization you have access to in this MCP session
switch_tenant	Change which Okareo organization subsequent tool calls operate against

Documentation & Templates

Tool	Description
get_docs	Query the Okareo documentation system for conceptual or user-legible explanations
get_templates	Retrieve prompt templates for common Okareo patterns (works offline)
get_reps_baseline	Serve REPS agent-evaluation baseline material (scenario banks, drivers, checks, eval configs) from the latest tagged okareo-tools release

---

Troubleshooting

Symptom	Cause	Fix
okareo-mcp: command not found	Not installed or not in PATH	Run pip install -e . (dev) or use uvx okareo-mcp (user)
Server exits with API key error	OKAREO_API_KEY not set	Export it: export OKAREO_API_KEY="..."
pip install fails on Python 3.13+	Okareo SDK requires Python <3.13	Use Python 3.10–3.12
Copilot can't connect (Docker)	Wrong URL	Ensure URL ends with /sse and port matches
Cursor doesn't pick up API key	Cursor launched from Dock, not terminal	Launch Cursor from terminal: cursor .

---

Contributing

This repository is a curated public mirror; the canonical source is maintained by Okareo. We welcome issues and consider community pull requests — see CONTRIBUTING.md for how proposed changes are reviewed and ported.

License & Trademarks

The Okareo MCP server source code is licensed under the Apache License 2.0 (see also NOTICE).

"Okareo", the Okareo logo, and Okareo product names are trademarks of Okareo, Inc. and are not covered by the Apache 2.0 license. See TRADEMARK.md for permitted use.