klavis

Implements an intelligent MCP router that dynamically exposes tools to AI agents in stages based on context relevance, preventing context window overload by avoiding simultaneous exposure to hundreds of tools. Uses a progressive discovery pattern where tools are surfaced incrementally as the agent's conversation evolves, with schema-based tool filtering and relevance ranking to match agent intent to available capabilities across 50+ integrated services.

Manages 50+ production-ready MCP servers across diverse service categories (CRM, communication, databases, content platforms) with unified OAuth2 authentication flows and API key management. Each service has a dedicated MCP server implementation (Python, TypeScript, or Go) that handles service-specific authentication patterns, token refresh, and credential storage, all coordinated through a central Management API that provisions and configures servers at runtime.

Provides specialized MCP servers for CRM and sales platforms with support for service-specific features like SOQL queries (Salesforce), deal pipeline management (HubSpot), task automation (Asana), and relationship mapping (Affinity). Each server implements authentication patterns specific to the platform, handles pagination and rate limits, and exposes domain-specific operations (e.g., creating opportunities, updating deal stages, managing contacts).

Provides MCP servers for communication and content platforms with support for message sending, channel management, user interaction, and content publishing. Includes Slack message posting with formatting, Discord bot integration, email sending via Resend, and WordPress content management, each with platform-specific authentication and rate limiting.

Provides MCP servers for database operations and web scraping with support for SQL queries, connection pooling, and structured data extraction from web pages. Includes servers for common databases (PostgreSQL, MySQL, MongoDB) and web scraping tools (Brave Search, Tavily, Exa) with built-in pagination, result formatting, and error handling.

Provides MCP servers for content and productivity platforms with support for video metadata retrieval (YouTube), document management (Google Docs/Sheets), note-taking (Notion), and database operations (Airtable). Each server implements platform-specific authentication, pagination, and data transformation to expose content operations through consistent MCP interfaces.

Provides MCP servers for specialized search and research APIs with support for semantic search, web search, and research-focused result ranking. Includes Tavily (research-optimized search), Exa (semantic search), and Brave Search (privacy-focused search), each with result ranking, snippet extraction, and pagination support optimized for agent-based research workflows.

Provides a production Go-based MCP server for GitHub with comprehensive repository operations including code search, pull request management, issue tracking, and workflow automation. Implements GitHub-specific patterns like branch protection rules, status checks, and webhook management, with native Go performance optimizations and concurrent API request handling.

Automatically generates Python and TypeScript SDKs from OpenAPI specifications and MCP server schemas, enabling developers to interact with Klavis APIs and MCP servers through native language bindings rather than raw HTTP calls. The pipeline uses schema introspection to generate type-safe client code with automatic request/response serialization, error handling, and async/await support, reducing boilerplate and improving developer experience across language ecosystems.

Provides open-source Strata deployment as containerized services that can run on-premises or in private cloud environments, with configuration-driven setup for MCP server registration, tool discovery rules, and authentication backends. Developers define Strata behavior through YAML/JSON configuration files that specify which MCP servers to load, tool filtering rules, and relevance scoring parameters, then deploy via Docker Compose or Kubernetes without modifying source code.

Bridges Klavis MCP tools to OpenAI's function calling API by automatically converting MCP tool schemas to OpenAI function definitions and mapping function call responses back to MCP tool invocations. Handles the schema translation layer transparently, allowing OpenAI models to invoke Klavis tools through native function calling without agents needing to understand MCP protocol details, including automatic parameter validation and error handling.

Exposes REST endpoints for programmatic MCP server management, allowing developers to create, configure, authenticate, and destroy MCP server instances at runtime through API calls rather than manual configuration. Includes endpoints for setting authentication credentials, querying available tools, testing connections, and monitoring server health, enabling dynamic agent infrastructure that can provision tool access on-demand.

Exposes REST endpoints for invoking MCP tools at runtime, handling parameter marshaling, error responses, and result formatting transparently. Developers call a single `/tools/invoke` endpoint with tool name and parameters, and the API routes the request to the appropriate MCP server, executes the tool, and returns structured results with error details if invocation fails, abstracting away MCP protocol complexity.

Manages OAuth2 tokens for all integrated services, handling initial authorization flows, token storage, automatic refresh before expiration, and revocation. Implements secure credential storage with encryption at rest, automatic refresh scheduling that proactively refreshes tokens before expiration to prevent tool invocation failures, and fallback mechanisms for refresh failures.

Provides REST endpoints for managing users, organizations, and access control in multi-tenant deployments, enabling partners to build branded agent platforms on top of Klavis infrastructure. Supports user creation, role-based access control (RBAC), organization isolation, and usage tracking per user/org, allowing partners to implement their own billing and access policies without modifying Klavis core.

Provides 50+ production-ready MCP server implementations (Python, TypeScript, Go) for major platforms (Salesforce, Slack, GitHub, Notion, HubSpot, Airtable, etc.), each implementing service-specific API patterns, error handling, and optimization strategies rather than generic REST-to-MCP adapters. Each server is maintained separately with service-specific tests, documentation, and version compatibility tracking.