$0 Forever
For individuals & small teams exploring Okta automation.
- Full read access to Okta APIs
- Users, Groups, Apps, Policies, Logs
- Basic MCP & Agent Mode
- CSV export & OAuth2 auth
Built on Anthropic's Code Execution Pattern Stop filling your context window with raw JSON. Tako MCP autonomously reasons through your Okta environment and executes server-side code to deliver only the results that matter.
- Dual-Mode: STDIO & HTTP Transports
- 90% Token Reduction via Code Execution
- Zero Result Limits for Thousands of Entities
Live Demo
Tako MCP in Claude Desktop
Watch the agent process thousands of users server-side without context window limitations.
Core Logic
Deferred Tool Discovery.
Tako MCP executes server-side code instead of streaming raw tool data, preserving your context window for complex reasoning.
Raw Data to LLM
1
User asks "Find users without MFA"
2
Server streams 5,000 users as raw JSON
💥
Context window exceeded. Results truncated.
Code to Sandbox
1
LLM receives 3 sample rows + API schema
2
LLM writes Python analysis script
[okta_audit.py]
✓
Sandbox executes script. Returns 100% of data via CSV.
Architectural Advantages
90% LESS TOKENS
Only sample data and metadata go to the LLM, keeping costs low and speed high.
NO RESULT LIMITS
Process 10,000+ entities without hitting token limits or truncation errors.
Secure Sandbox
Code execution is isolated in an AST-based sandbox. Read-only by design.
Configuration
Two Powerful Operating Modes.
Choose the interaction pattern that fits your workflow—from standard tool access to fully autonomous agent logic.
Default MCP Mode
Claude & Cursor
Direct tool access for standard IDE assistants.
Full Toolset
All 20+ Okta tools loaded into context immediately.
Sample Arrays
Returns metadata and sample results directly to LLM.
Autonomous Agent Mode
Progressive
Lower initial tokens; tools discovered as needed.
Full Scale
Processes thousands of rows via server-side code.
Agent Frameworks
LangGraph, LlamaIndex, and complex autonomous loops.
Connectivity
Works Everywhere.
Native STDIO for local desktop environments and HTTP/SSE for remote enterprise deployments.
Desktop & IDEs
Claude Desktop, Cursor AI, VS Code, Zed, and Windsurf.
Agent Frameworks
LangGraph, LlamaIndex, Goose, and custom autonomous agents.
Enterprise Gateways
Obot.ai, Smithery.ai, Lasso Security, and custom clients.
Core Capabilities
Built for the Enterprise.
Production-ready MCP server with enterprise-grade security, flexible transport options, and deterministic execution.
Dual-Mode Operation
Switch between standard MCP mode for direct tool access or autonomous agent mode for complex logic execution.
Flexible Transports
Native STDIO for desktop clients like Claude or Cursor, and HTTP/SSE for remote enterprise team deployments.
Enterprise Auth
OAuth2 with Private Key JWT (RFC 7523) for production or simple API tokens for development.
CSV Export
Automatic large-dataset export to files to reduce token costs and keep context windows clean.
Open Standards
🌟 Elastic License 2.0
📐 MCP Specification
🏗️ Production Ready
🔐 OAuth2 RFC 7523
Enterprise Security Sandbox
AST-based sandbox with multi-layer protection designed for safe code execution in production environments.
PATTERN BLOCKING
Dangerous operations and OS-level calls are blocked at the Abstract Syntax Tree level before execution.
IMPORT CONTROL
Pre-injected modules only. No arbitrary third-party imports or unauthorized library access allowed.
FUNCTION WHITELIST
Strict enforcement of safe, built-in functions. Isolated logic ensures zero data exfiltration.
HTTP RESTRICTION
Read-only GET request enforcement ensures your identity data is analyzed but never modified.
Pricing
Simple, transparent pricing.
Start free with full read access. Upgrade when you need write operations and enterprise controls.
Custom Coming Soon
For organizations at scale needing write operations and advanced controls.
- Write operations & approvals
- Multi-tenant support & RBAC
- Custom tool builder
- SSO & audit log streaming
- Priority support & SLA
Onboarding
Ready in 5 Minutes.
Three simple steps to start querying Okta with natural language using Tako MCP.
Clone Repository
Download the source code. Use Docker for quick setup or Python for more control.
git clone github.com/fctr-id/fctr-okta-mcp-server
Configure Access
Add your Okta org URL and choose API Token or OAuth2 for authentication.
OKTA_CLIENT_ORGURL=https://your-org.okta.com
Connect & Query
Add the MCP server to Claude Desktop or Cursor and start querying.
"Find all users without MFA" ✓
Get Started
Start querying Okta with AI.
Free, open source, self-hosted.
Deploy in 5 minutes and eliminate the context window limits of traditional MCP servers.
MIT Licensed · Production Ready · Community Support
FAQ
Commonly asked questions.
No, this is an independent open-source project built by Fctr Identity. We are not affiliated with Okta, Inc.
Yes. The server is read-only by default, runs in a sandboxed environment, and all code is validated before execution. Your credentials never leave your infrastructure.
Any MCP-compatible client: Claude Desktop, VS Code, Cursor, Zed, and agent frameworks like LangGraph and LlamaIndex.
Absolutely. The community edition is fully self-hosted. Run it on your laptop, in Docker, or on your own servers.
Works with all Okta editions: Workforce Identity, Customer Identity (Auth0), and free Developer accounts.
Model Context Protocol (MCP) is an open standard that connects AI assistants to external tools and data sources. Tako MCP Server implements Anthropic's Code Execution with MCP pattern (deferred tool discovery), allowing your AI to discover Okta APIs progressively and generate Python code for querying your environment.
Tako MCP Server supports both API Token (simple setup) and OAuth2 with Private Key JWT (recommended for production). OAuth2 provides scoped permissions and better security for enterprise deployments.
No, Docker is optional. You can run Tako MCP Server locally with Python (we recommend uv for package management), but Docker provides the fastest setup without managing dependencies.
Basic Mode exposes 20+ Okta tools directly to the AI, using more context upfront. Agent Mode (also called deferred tool discovery) uses meta-tools for progressive discovery, saving tokens and enabling autonomous workflows. Agent Mode is ideal for complex queries and agent frameworks like LangGraph.
Tako MCP Server includes configurable concurrent request limits based on your Okta plan (Integrator, One App, Enterprise, or Workforce Identity). The server monitors rate limits and provides warnings if limits are exceeded.
The open-source community edition is read-only by design. Write operations, approval workflows, and role-based access control are available in the upcoming Enterprise Edition.