Content
# clawconnect
```
████████████████████████████████████████
██ ██
██ ░░ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ ░░ ██
██ ░░ ██ ╔══╗ CLAW ██ ░░ ██
██ ░░ ██ ║██║ CONNECT ██ ░░ ██
██ ░░ ██ ╚══╝ ██ ░░ ██
██ ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ ██
████████████████████████████████████████
CONSUME. CONNECT. CONTROL.
```
**Cowork-style connectors for OpenClaw agents — via CLI, Telegram, and MCP.**
---
## What is clawconnect?
[Claude Cowork](https://claude.ai) has a polished connector store — authenticate Gmail, Slack, Notion, and dozens of other services in a few clicks, with explicit permission controls (read-only, write, approval required). OpenClaw agents have no equivalent: connectors are assembled through plugins, environment variables, and manual config.
clawconnect closes that gap. It gives OpenClaw users a Cowork-style connector experience from the terminal, with three components that work together:
1. **CLI** — authenticate services, manage credentials, and execute actions directly
2. **MCP bridge** — exposes all connected services as MCP tools, readable by both OpenClaw and Cowork custom connector slots
3. **Telegram companion** — agents can request connector actions on your behalf; you approve via inline keyboard before anything executes
---
## How clawconnect differs from existing tools
Several open-source Claude Cowork alternatives exist — [OpenWork](https://openworklabs.com), [Eigent](https://eigent.ai), [open-claude-cowork](https://github.com/composio-community/open-claude-cowork) — but none of them solve the same problem:
| | OpenWork | Eigent | open-claude-cowork | **clawconnect** |
|---|---|---|---|---|
| Target | Cowork desktop replacement | Multi-agent workforce | Cowork desktop replacement | OpenClaw connector layer |
| Interface | Desktop GUI | Desktop GUI | Electron GUI | CLI + Telegram + MCP |
| OpenClaw integration | ✗ | ✗ | Partial (messaging only) | ✓ Native via MCP |
| Cowork bridge | ✗ | ✗ | ✗ | ✓ Custom connector URL |
| Agent-initiated actions | ✗ | ✗ | ✗ | ✓ With user approval gate |
| Telegram approval flow | ✗ | ✗ | ✓ (messaging only) | ✓ Full connector approval |
| No GUI required | ✓ | ✗ | ✗ | ✓ |
| Custom connectors | Via MCP | Via MCP | Via skills | ✓ YAML/JSON drop-in |
| Requires Anthropic account | ✗ | ✗ | ✗ | ✗ |
**The specific gap clawconnect fills:** OpenClaw agents can discover and call connector actions as MCP tools — and users can approve or deny those requests via Telegram — without any GUI, without Anthropic credentials, and without replacing their existing OpenClaw setup.
---
## Built-in Connectors
| Connector | Auth | Key Actions |
|---|---|---|
| `gmail` | OAuth2 | list, get, send, search messages |
| `google-drive` | OAuth2 | list, get, download, upload files |
| `calendar` | OAuth2 | list events, create event |
| `slack` | OAuth2 | list channels, send message, get messages |
| `notion` | API Key | databases, pages — list, query, create, update |
| `github` | API Key | repos, issues, PRs — list, create, get |
| `custom` | API Key | any MCP-compatible server URL |
---
## Install
```bash
pip install clawconnect
# Optional: YAML support for custom connectors
pip install pyyaml
# Or from source:
git clone https://github.com/terence-ma/clawconnect
cd clawconnect && pip install -e .
```
---
## Quick Start
```bash
# 1. Initial setup
clawconnect init
# 2. Browse connectors
clawconnect connectors list
# 3. Connect services
clawconnect connect gmail # OAuth2 — opens browser
clawconnect connect notion --api-key secret_xxx
clawconnect connect github --api-key ghp_xxx
# 4. Run an action
clawconnect run gmail list_messages --params '{"max_results": 5}'
clawconnect run slack send_message --params '{"channel": "#general", "text": "hello"}'
# 5. Start MCP bridge (OpenClaw + Cowork)
clawconnect bridge start
# 6. Start Telegram companion bot
clawconnect telegram start
```
---
## MCP Bridge
The bridge server exposes all connected services as MCP tools over HTTP.
One URL works in both OpenClaw and Cowork.
**OpenClaw** — add to `~/.openclaw/openclaw.json`:
```json
{
"mcp": {
"servers": {
"clawconnect": {
"url": "http://127.0.0.1:18800/mcp"
}
}
}
}
```
**Cowork** — Settings → Connectors → Add custom connector:
```
URL: http://127.0.0.1:18800/mcp
```
Your agents can then call:
- `gmail__send_message`, `gmail__list_messages`
- `slack__send_message`, `slack__list_channels`
- `notion__create_page`, `notion__query_database`
- `github__create_issue`, `github__list_repos`
- Any custom connector you've defined
---
## Telegram Companion
Agents can request connector actions via the bridge. The Telegram bot
surfaces those requests to you as inline keyboard approvals.
```
Agent wants to: gmail → send_message
To: someone@example.com
Subject: Weekly report
Body: Here is the summary...
[✅ Approve] [❌ Deny]
```
**Setup:**
1. Create a bot with [@BotFather](https://t.me/botfather)
2. `clawconnect init` → enter bot token
3. `clawconnect telegram start`
---
## Custom Connectors
### Method 1: Drop a YAML file (no Python required)
Create `~/.clawconnect/connectors/my-service.yaml`:
```yaml
name: my-service
display_name: My Service
description: Connect to My Service API
category: productivity
auth: api_key
actions:
- name: list_items
description: List all items
params:
limit: int
query: str
- name: create_item
description: Create a new item
params:
title: str
body: str
telegram_prompts:
create_item:
- field: title
prompt: "Item title:"
- field: body
prompt: "Item content:"
# Auto-execute HTTP calls without writing Python
executor:
base_url: https://api.my-service.com/v1
auth_header: Authorization
auth_prefix: Bearer
actions:
list_items:
method: GET
path: /items
query_params: [limit, query]
create_item:
method: POST
path: /items
body_params: [title, body]
```
That's it. No code. Restart clawconnect and `my-service` appears in the connector list.
### Method 2: Python executor (full control)
Add to `connectors/registry.py` and `connectors/executor.py` for
connectors with complex auth, pagination, or response transformation needs.
See existing connectors for the pattern.
---
## Architecture
```
clawconnect/
├── cli.py # CLI entry point (click + rich)
├── connectors/
│ ├── registry.py # Built-in connector definitions
│ ├── loader.py # YAML/JSON custom connector loader
│ ├── oauth.py # OAuth 2.0 browser flow
│ └── executor.py # API call implementations
├── bridge/
│ └── server.py # MCP JSON-RPC bridge server
└── telegram/
└── bot.py # Telegram approval companion
Config: ~/.clawconnect/
├── settings.json # Gateway URL, ports, bot token
├── tokens.json # Credentials (user-only permissions)
└── connectors/ # Drop custom YAML/JSON connectors here
└── example-connector.yaml.disabled
```
---
## Security
- Tokens stored at `~/.clawconnect/tokens.json` with user-only file permissions
- OAuth client secrets read from environment variables — never stored to disk
- Every agent-initiated action requires explicit Telegram approval before execution
- MCP bridge binds to `127.0.0.1` only — not exposed externally by default
- Expose via Cloudflare Tunnel if remote access needed (not recommended without auth)
---
## License
MIT — free to use, modify, and redistribute.
