Goose AI Agent

The Speech MCP server is an open-source, privacy-focused, and LLM-agnostic tool that allows for voice interaction with AI agents like Goose. It operates entirely locally, making it cost-effective and accessible without internet connectivity.

Key Capabilities

• Local Speech Processing: Utilizes Faster Whisper (speech-to-text) and Coqui TTS (text-to-speech).
• Voice Selection: Access to 54 different natural-sounding voices.
• Multi-Speaker Narration: Ability to generate and play conversations between multiple distinct voices.
• Audio Transcription: Converts audio/video content to text, including timestamps and speaker detection.
• System Integration: Enables voice commands to control applications (e.g., Discord), navigate the file system, and manage development environments.

OIDC Proxy is a Cloudflare Worker designed to authenticate GitHub Actions OIDC tokens and proxy requests to an upstream API by injecting a real API key. This architecture allows CI workflows to call sensitive APIs without the need to store long-lived secrets (like API keys) directly in GitHub.

Workflow Flow: GitHub Actions (OIDC token) → Worker (validate JWT, inject API key) → Upstream API

1. A GitHub Actions workflow generates an OIDC token with a specific audience.
2. The workflow sends requests to the proxy, using the OIDC token as the authentication credential.
3. The worker validates the JWT against GitHub's JWKS, checking the issuer, audience, token age, and repository permissions.
4. Upon successful validation, the worker forwards the request to the upstream API with the actual API key injected into the appropriate header.

Goose is an open-source, MCP-compatible AI agent available as both a CLI and a desktop application. It uses the Model Context Protocol (MCP) to interact with APIs, tools, and data systems through a common, deterministic interface. This allows the agent to call tools without guesswork, focusing on executing tasks rather than interpreting API structures.

Key features include:

• Model-agnosticism: Works with various LLMs (e.g., Claude, OpenAI) via secure endpoints.
• Tool-agnosticism: Connects to internal APIs, open-source tools, and SaaS products via MCP servers.
• Security controls: Supports defining which models can invoke specific tools and annotating tools as read-only or destructive to trigger user confirmation.

codename goose is an open-source, on-machine AI agent designed to automate engineering tasks and improve productivity. It functions as an autonomous assistant that can execute complex instructions by coordinating several core components:

• LLM Providers: Supports a wide range of Large Language Models (LLMs) to power the agent's reasoning.
• Extensions: Uses the Model Context Protocol (MCP) to connect with external tools and applications (e.g., GitHub, Google Drive, JetBrains IDEs). This allows you to bring your own custom integrations.
• Interfaces: Available as both a Desktop application and a Command-Line Interface (CLI), sharing the same configuration across both platforms.

The documentation/automation/ directory contains automated pipelines designed to synchronize goose documentation with code changes. These projects use a hybrid approach combining deterministic shell scripts for extraction/comparison and AI-powered recipes for synthesis and updates.

Currently planned automation projects include:

• cli-command-tracking: Tracks CLI commands and options to update CLI documentation.
• provider-tracking: Tracks supported AI providers to update provider documentation.
• extension-tracking: Tracks built-in extensions to update extension documentation.

The goose-sdk crate serves as the bindings layer for Goose. It provides a cross-language version of the Goose API and contains the shared types used for both ACP (Agent Communication Protocol) and SDK access.

When compiled with the --features uniffi flag, it generates native bindings for:

• Python: Namespace goose
• Kotlin: Namespace io.github.aaif_goose

Through the UniFFI surface, developers can:

• Construct providers
• Stream provider completions
• Perform non-streaming completions
• Pass rich message and tool content across the FFI boundary.

The Firecrawl MCP server provides the following web-based capabilities to your AI agents:

• Single Page Scraping: Extract clean, structured content from individual web pages.
• Batch Processing: Scrape multiple URLs efficiently using batch operations.
• Site Crawling: Crawl entire websites or specific sections with customizable depth.
• Search Integration: Search across crawled content with advanced filtering.

Goose provides a wide range of capabilities through various extensions, regardless of whether you are using the Desktop App or the CLI. Core capabilities include:

• Development & Code: Editing files, running shell commands, working with Git (via GitHub tools), compiling/testing Rust projects, and working with TypeScript/Electron apps.
• Memory & Context: Storing and retrieving preferences, project details, and workflows globally or locally per project.
• Web & Data Processing: Web scraping, API interaction, and processing file formats like PDF, DOCX, and Excel.
• System Control (macOS): Taking screenshots, analyzing the screen, and controlling applications via AppleScript.
• Task Management: Creating scheduled tasks and spawning specialized subagents for complex workflows.
• Extensions: Searching for and enabling additional specialized tools and API integrations.

The transition to the ACP-based architecture is being rolled out in phases: