# Tedix [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/tedix-hq/tedix  
**GitHub Stars:** 14  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/tedix

## Description
Open-source AI coworkers that show their work: approval before risky steps, a record of every run.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What Tedix MCP server does

The Tedix MCP server exposes Tedix’s AI coworker workflow to compatible clients through the MCP gateway. Tedix calls its AI coworkers “tedis”; each can receive recurring work and operate with its own memory, skills, and policies. The platform is designed for jobs where an agent should not act without appropriate review.

A request can be handled directly in Home or passed to a tedi. Work Items track the assigned owner, risk level, and budget. When a step is considered risky, the work pauses for approval. The requester cannot approve its own request. After execution, Tedix stores a record of the worker, tools used, approval, and result.

## How it works

The Tedix MCP server sits between an MCP-compatible client and Tedix’s application platform. The README identifies Claude, ChatGPT Work, and Codex integrations through the Tedix plugin, and points other MCP-compatible clients to the MCP gateway.

A typical flow starts with a job submitted in Home or by a connected client. Tedix either responds directly or assigns the request to a tedi. The tedi works within the job’s ownership, risk, and budget controls. Approval gates apply before risky actions, while the completed run remains available as an operational record.

The project is built on Cloudflare Workers, Durable Objects, Workflows, D1, and R2. The README does not provide a separate list of individual MCP method names, so capability descriptions here refer to the documented product concepts rather than undocumented tool identifiers.

## Setup and configuration

For local use, install Git, Bun, and Node.js 22 or newer. The README specifically requires real Node.js rather than a Bun shim. Clone the repository, enter it, and run `bun run-local`. The launcher starts the API, database, and web application, then reports a local address, normally `http://localhost:3030`.

On first launch, name a local organization. Tedix creates the organization, an owner account, and an initial tedi. Local data is stored in `.wrangler/run-local`. Model calls are disabled by default. To enable model turns through Workers AI, log in with Wrangler and provide the Cloudflare account ID using the `--workers-ai-account` option. The `--smoke` option checks onboarding and exits; `--demo` loads sample data.

Tedix Cloud is an invited beta. Cloud users can connect Claude chat or Cowork through a connector URL, use the Tedix plugin with supported products, or install the CLI and authenticate with `tedix login`.

## Tools and capabilities

The Tedix MCP server supports these documented capabilities:

- Submit or route work to a tedi with its own memory, skills, and policies.
- Represent work with an owner, risk level, and budget.
- Pause risky steps until a human approval is recorded.
- Preserve a run record containing the worker, tools used, approval, and result.
- Support recurring work for teams using local Tedix or the invited cloud service.

These controls suit supplier review, customer-impacting actions, money-related tasks, and production operations when the workflow requires a visible approval step and a record of what happened.

## Limitations and notes

Tedix is marked beta in the repository. Cloud access is invitation-based rather than generally available. Local mode does not make model calls unless Workers AI is enabled with a paid Cloudflare account.

The repository states that licensing is mostly AGPL-3.0, with some packages under Apache-2.0 or MIT; consult its licensing files when redistributing or modifying the project. The CLI sends no telemetry according to the README. The provided material does not specify individual MCP tool names, transport details, or a complete compatibility matrix beyond the named integrations and the general MCP gateway.

_Full upstream README: https://allmcps.com/mcp/tedix/readme_