Skip to content

Commit 22ac392

Browse files
abrichrclaude
andcommitted
Add Anthropic ComputerUse integration information
- Add detailed comparison of OmniMCP and Anthropic ComputerUse approaches - Describe key architectural differences and integration opportunities - Add TODO comment for future ComputerUse integration possibilities 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
1 parent e31a8dd commit 22ac392

File tree

2 files changed

+34
-2
lines changed

2 files changed

+34
-2
lines changed

omnimcp/README.md

Lines changed: 26 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -95,9 +95,33 @@ OmniMCP requires access to an OmniParser server for analyzing screenshots:
9595

9696
By default, OmniMCP will fail if it can't connect to an OmniParser server. Use the `--allow-no-parser` flag to run with limited functionality when no parser is available.
9797

98-
### TODO: Anthropic ComputerUse Integration
98+
### Future Direction: Anthropic ComputerUse Integration
9999

100-
Future versions may integrate with Anthropic's ComputerUse system, which provides an official way for Claude to interact with computers through a virtualized desktop environment.
100+
OmniMCP and Anthropic's [ComputerUse](https://docs.anthropic.com/en/docs/agents-and-tools/computer-use) both enable Claude to control computers, but with different architectural approaches:
101+
102+
#### Key Differences
103+
104+
**Integration Approach:**
105+
- **OmniMCP** uses OmniParser for understanding UI elements
106+
- **ComputerUse** captures screenshots and provides them directly to Claude
107+
108+
**Environment:**
109+
- **OmniMCP** runs directly on the host system with minimal dependencies
110+
- **ComputerUse** operates in a containerized virtual desktop environment
111+
112+
**MCP vs. Anthropic-defined Tools:**
113+
- **OmniMCP** uses the Model Control Protocol (MCP), a structured protocol for AI models to interact with tools
114+
- **ComputerUse** uses Anthropic-defined tools (`computer`, `text_editor`, and `bash`) via Claude's tool use API
115+
116+
#### Potential Integration Paths
117+
118+
Future OmniMCP development could:
119+
1. **Dual Protocol Support**: Support both MCP and Anthropic-defined tools
120+
2. **Container Option**: Provide a containerized deployment similar to ComputerUse
121+
3. **Unified Approach**: Create a bridge between MCP and ComputerUse tools
122+
4. **Feature Parity**: Incorporate ComputerUse capabilities while maintaining MCP compatibility
123+
124+
Both approaches have merits, and integrating aspects of ComputerUse could enhance OmniMCP's capabilities while preserving its lightweight nature and existing MCP integration.
101125

102126
## Features
103127

openadapt/run_omnimcp.py

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -50,6 +50,14 @@
5050
from openadapt.omnimcp import OmniMCP
5151
from openadapt.custom_logger import logger
5252

53+
# TODO: Consider Anthropic ComputerUse integration
54+
# Anthropic's ComputerUse (https://docs.anthropic.com/en/docs/agents-and-tools/computer-use)
55+
# provides an official approach for Claude to control computers. While OmniMCP already
56+
# implements a similar agent loop pattern, future work could:
57+
# 1. Refine our existing agent loop to better align with ComputerUse's approach
58+
# 2. Support Anthropic's containerized environment as a deployment option
59+
# 3. Offer compatibility with Anthropic's official computer control tools
60+
5361

5462
class OmniMCPRunner:
5563
"""OmniMCP runner with different modes of operation."""

0 commit comments

Comments
 (0)