Gget MCP

1# gget-mcp
2 
3[![Tests](https://github.com/longevity-genie/gget-mcp/actions/workflows/test.yml/badge.svg)](https://github.com/longevity-genie/gget-mcp/actions/workflows/test.yml)
4[![PyPI version](https://badge.fury.io/py/gget-mcp.svg)](https://badge.fury.io/py/gget-mcp)
5[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
6[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
7[![Code style: ruff](https://img.shields.io/badge/code%20style-ruff-000000.svg)](https://github.com/astral-sh/ruff)
8 
9MCP (Model Context Protocol) server for the [gget](https://github.com/pachterlab/gget) bioinformatics library.
10 
11![gget-mcp Example](images/screenshot_example.png)
12 
13This server implements the Model Context Protocol (MCP) for gget, providing a standardized interface for accessing powerful bioinformatics tools and databases. MCP enables AI assistants and agents to perform complex genomics queries through structured interfaces.
14 
15The gget bioinformatics toolkit provides efficient querying of genomic databases and includes functions for:
16- Gene and sequence information retrieval
17- BLAST and sequence alignment
18- Protein structure prediction
19- Gene expression analysis
20- Functional enrichment analysis
21- And much more!
22 
23## 🚀 Quick Example: What Can You Do?
24 
25<details>
26<summary>See how the gget MCP server works in practice with AI assistants</summary>
27 
28**👤 User:** "Analyze the TP53 gene - find its details and protein sequence, then check for cancer mutations"
29 
30**🤖 AI Assistant:** I'll help you analyze TP53 step by step:
31 
32*[Uses gget_search]* → Found TP53 (ENSG00000141510) on chromosome 17  
33*[Uses gget_info]* → TP53 is a protein-coding tumor suppressor gene  
34*[Uses gget_seq with translate=True]* → Retrieved 393 amino acid protein sequence  
35*[Uses gget_cosmic]* → Found hotspot mutations R175H, R248Q, R273H in multiple cancers
36 
37**👤 User:** "Now predict the 3D structure of this protein sequence: MSKGEELFTGVVPILVEL..."
38 
39**🤖 AI Assistant:** Let me analyze this sequence:
40 
41*[Uses gget_blast]* → Identified as Green Fluorescent Protein (GFP)  
42*[Uses gget_alphafold]* → Predicted classic beta-barrel structure with 90%+ confidence
43 
44**👤 User:** "Perform pathway enrichment for DNA repair genes: TP53, BRCA1, BRCA2, ATM"
45 
46**🤖 AI Assistant:** *[Uses gget_enrichr]* → Top pathways: DNA Damage Response, Homologous Recombination, Cell Cycle Checkpoints, p53 Signaling
47 
48**See how powerful this is?** You can now ask natural language questions about genes, sequences, mutations, and pathways - the AI does the complex bioinformatics work for you!
49 
50</details>
51 
52## About MCP (Model Context Protocol)
53 
54<details>
55<summary>Learn more about the Model Context Protocol</summary>
56 
57MCP is a protocol that bridges the gap between AI systems and specialized domain tools. It enables:
58 
59- **Structured Access**: Direct connection to powerful bioinformatics tools and databases
60- **Natural Language Queries**: Simplified interaction with complex genomics workflows
61- **Type Safety**: Strong typing and validation through FastMCP
62- **AI Integration**: Seamless integration with AI assistants and agents
63 
64If you want to understand more about what the Model Context Protocol is and how to use it more efficiently, you can take the [DeepLearning AI Course](https://www.deeplearning.ai/short-courses/mcp-build-rich-context-ai-apps-with-anthropic/) or search for MCP videos on YouTube.
65 
66</details>
67 
68## Available Tools
69 
70<details>
71<summary>Comprehensive bioinformatics functionality through gget</summary>
72 
73### Gene Information & Search
74- **`gget_search`**: Find Ensembl IDs associated with search terms
75- **`gget_info`**: Fetch detailed information for Ensembl IDs
76- **`gget_seq`**: Retrieve nucleotide or amino acid sequences
77- **`gget_ref`**: Get reference genome information from Ensembl
78 
79### Sequence Analysis
80- **`gget_blast`**: BLAST nucleotide or amino acid sequences
81- **`gget_blat`**: Find genomic locations of sequences
82- **`gget_muscle`**: Align multiple sequences
83 
84### Expression & Functional Analysis
85- **`gget_archs4`**: Get gene expression data from ARCHS4
86- **`gget_enrichr`**: Perform gene set enrichment analysis
87 
88### Protein Structure & Function
89- **`gget_pdb`**: Fetch protein structure data from PDB
90- **`gget_alphafold`**: Predict protein structure using AlphaFold
91 
92### Cancer & Mutation Analysis
93- **`gget_cosmic`**: Search COSMIC database for cancer mutations
94 
95### Single-cell Analysis
96- **`gget_cellxgene`**: Query single-cell RNA-seq data from CellxGene
97 
98</details>
99 
100## Quick Start
101 
102<details>
103<summary>Installing uv (optional - uvx can auto-install)</summary>
104 
105```bash
106# Download and install uv
107curl -LsSf https://astral.sh/uv/install.sh | sh
108 
109# Verify installation
110uv --version
111uvx --version
112```
113 
114uvx is a very nice tool that can run a python package installing it if needed.
115 
116</details>
117 
118### Running with uvx
119 
120You can run the gget-mcp server directly using uvx without cloning the repository:
121 
122```bash
123# Run the server in HTTP mode (default)
124uvx gget-mcp http
125```
126 
127<details>
128<summary>Other uvx modes (STDIO, HTTP, SSE)</summary>
129 
130#### STDIO Mode (for MCP clients that require stdio)
131 
132```bash
133# Run the server in stdio mode
134uvx gget-mcp stdio
135```
136 
137#### HTTP Mode (Web Server)
138```bash
139# Run the server in streamable HTTP mode on default (3002) port
140uvx gget-mcp http
141 
142# Run on a specific port
143uvx gget-mcp http --port 8000
144```
145 
146#### SSE Mode (Server-Sent Events)
147```bash
148# Run the server in SSE mode
149uvx gget-mcp sse
150```
151 
152</details>
153 
154In cases when there are problems with uvx often they can be caused by cleaning uv cache:
155```
156uv cache clean
157```
158 
159The HTTP mode will start a web server that you can access at `http://localhost:3002/mcp` (with documentation at `http://localhost:3002/docs`). The STDIO mode is designed for MCP clients that communicate via standard input/output, while SSE mode uses Server-Sent Events for real-time communication.
160 
161**Note:** Currently, we do not have a Swagger/OpenAPI interface, so accessing the server directly in your browser will not show much useful information. To explore the available tools and capabilities, you should either use the MCP Inspector (see below) or connect through an MCP client to see the available tools.
162 
163## Configuring your AI Client (Anthropic Claude Desktop, Cursor, Windsurf, etc.)
164 
165We provide preconfigured JSON files for different use cases. Here are the actual configuration examples:
166 
167### STDIO Mode Configuration (Recommended)
168 
169Use this configuration for most AI clients. Use this mode when you want to save large output files (sequences, structures, alignments) to disk instead of returning them as text.
170Create or update your MCP configuration file:
171 
172```json
173{
174  "mcpServers": {
175    "gget-mcp": {
176      "command": "uvx",
177      "args": ["--from", "gget-mcp@latest", "stdio"]
178    }
179  }
180}
181```
182 
183### HTTP Mode Configuration
184 
185For HTTP mode:
186 
187```json
188{
189  "mcpServers": {
190    "gget-mcp": {
191      "command": "uvx",
192      "args": ["--from", "gget-mcp@latest", "server"]
193    }
194  }
195}
196```
197 
198### Configuration Video Tutorial
199 
200For a visual guide on how to configure MCP servers with AI clients, check out our [configuration tutorial video](https://www.youtube.com/watch?v=Xo0sHWGJvE0) for our sister MCP server (biothings-mcp). The configuration principles are exactly the same for the gget MCP server - just use the appropriate JSON configuration files provided above.
201 
202### Inspecting gget MCP server
203 
204<details>
205<summary>Using MCP Inspector to explore server capabilities</summary>
206 
207If you want to inspect the methods provided by the MCP server, use npx (you may need to install nodejs and npm):
208 
209For STDIO mode with uvx:
210```bash
211npx @modelcontextprotocol/inspector --config mcp-config.json --server gget-mcp
212```
213 
214You can also run the inspector manually and configure it through the interface:
215```bash
216npx @modelcontextprotocol/inspector
217```
218 
219After that you can explore the tools and resources with MCP Inspector at which is usually at 6274 port (note, if you run inspector several times it can change port)
220 
221</details>
222 
223### Integration with AI Systems
224 
225Simply point your AI client (like Cursor, Windsurf, ClaudeDesktop, VS Code with Copilot, or [others](https://github.com/punkpeye/awesome-mcp-clients)) to use the appropriate configuration file from the repository.
226 
227## Platform-Specific Setup Guides
228 
229### Claude Desktop + gget_mcp Step-by-Step Guide (Windows)
230 
231<details>
232<summary>Comprehensive Windows setup guide with Google Drive integration</summary>
233 
234#### Overview
235This guide will walk you through setting up Claude Desktop with the gget_mcp extension and Google Drive integration on Windows. By the end, you'll be able to use Claude to fetch biological data (like gene sequences) and save them directly to your Google Drive folder with offline access.
236 
237#### Prerequisites
238- Windows PC with administrator access
239- Google account
240- Stable internet connection
241 
242#### Step 1: Set Up Google Drive for Desktop with Offline Access
243 
2441. Download and install [Google Drive for Desktop](https://www.google.com/drive/download/)
2452. Launch the application and sign in with your user account
2463. Connect your project's shared Google account (if applicable)
2474. **Configure offline access for your working folder:**
248   - Navigate to the folder you want to work with in Google Drive
249   - Right-click on the folder (e.g., "work" or your project folder)
250   - Select **"Available offline"** from the context menu
251   - This makes the folder accessible at a path like `C:\GDrive\holy-bio-mcp\My Disk\work`
252 
253   **Important:** This is different from just syncing - offline access ensures the files are locally available while still being part of your Google Drive structure.
254 
255#### Step 2: Install Claude Desktop
256 
2571. Download [Claude Desktop](https://claude.ai/download) for Windows
2582. Run the installer and follow the setup wizard
2593. Sign in with your Anthropic/Claude account
2604. Complete the initial setup
261 
262#### Step 3: Install uv Package Manager
263 
2641. Install `uv` using one of these methods:
265 
266   **Option A: Using winstall**
267   - Visit [https://winstall.app/apps/astral-sh.uv](https://winstall.app/apps/astral-sh.uv)
268   - Follow installation instructions
269 
270   **Option B: Using PowerShell**
271   ```powershell
272   powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
273   ```
274 
2752. Restart your terminal after installation
276 
277#### Step 4: Install Node.js
278 
279You can install Node.js using either method:
280 
281**Option A: Native Installer**
2821. Visit [https://nodejs.org/en/download](https://nodejs.org/en/download)
2832. Download the Windows installer (LTS version recommended)
2843. Run the installer with default settings
285 
286**Option B: Using Chocolatey**
287```powershell
288choco install nodejs
289```
290 
291Both methods work equally well - choose whichever you prefer.
292 
293#### Step 5: Verify Installation & Cache Packages
294 
295Open Command Prompt or PowerShell and run:
296 
297```bash
298uvx --from gget-mcp@latest stdio
299```
300 
301**Expected behavior:**
302- The command should start without errors
303- Packages will be downloaded and cached
304- If it runs successfully, everything is configured correctly
305- You can press `Ctrl+C` to stop it
306 
307#### Step 6: Configure Claude Desktop
308 
309##### 6.1: Enable Filesystem Extension
310 
3111. Open Claude Desktop
3122. Go to **Settings** (gear icon)
3133. Navigate to the **"Extensions"** tab in the left sidebar
3144. Enable the **Filesystem** Extension
3155. Add your Google Drive offline folder to the allowed directories (e.g., `C:\GDrive\holy-bio-mcp\My Disk\work`)
316 
317**Note:** The Filesystem extension is built into Claude Desktop and uses MCP under the hood, but it's managed directly by the application. You don't need to configure it in the JSON file.
318 
319##### 6.2: Add gget_mcp Server
320 
3211. In the same manner, go to **Settings**, navigate to the **"Developer"** tab in the left sidebar
3222. In **Developer** settings tab, click **"Edit Config"** button
3233. This will open the `claude_desktop_config.json` file
3244. Edit it and add the gget_mcp server configuration:
325 
326```json
327{
328  "mcpServers": {
329    "gget": {
330      "command": "uvx",
331      "args": [
332        "--from",
333        "gget-mcp@latest",
334        "stdio"
335      ]
336    }
337  }
338}
339```
340 
3415. Save the configuration file
342 
343#### Step 7: Restart Claude Desktop
344 
3451. Completely close Claude Desktop (check system tray)
3462. Restart the application
3473. **Note:** On first launch, you might experience a timeout - this is normal
3484. If needed, restart Claude Desktop a second time
349 
350#### Step 8: Verify Extensions Are Active
351 
3521. Open Claude Desktop Settings
3532. Go to the **Developer** tab
3543. Confirm that both components are enabled:
355   - ✅ **Filesystem Extension** - should show your configured Google Drive path and show "Running" status
356   - ✅ **gget MCP Server** - should show "Running" status
357 
358#### Step 9: Test the Setup
359 
360Now let's verify everything works! In Claude Desktop, create a new chat.
361Click "Search and tools" icon, select and enable both gget_mcp and filesystem
362 
363Try this prompt:
364 
365```text
366Please search for the COL1A1 gene, retrieve its protein sequence, 
367and save it to C:\GDrive\holy-bio-mcp\My Disk\work\test_sequences\ as a FASTA file.
368```
369 
370**Expected result:**
371- Claude will use gget_mcp to search for the gene
372- It will fetch the sequence data
373- It will save the file to your specified folder
374- The file will sync to Google Drive automatically
375- Claude will be able to read it using Filesystem Extension.
376 
377#### Troubleshooting
378 
379##### MCP Server Not Connecting
380 
381- Ensure `uv` and `node` are in your system PATH
382- Try running the verification command again from Step 5
383- Restart Claude Desktop completely
384 
385##### Filesystem Permission Errors
386 
387- Check that the Filesystem Extension has access to your Google Drive folder
388- Verify the folder has offline access enabled in Google Drive
389- Try running Claude Desktop as administrator
390- Ensure the path is correct (use the exact path shown in File Explorer)
391 
392##### Timeout on First Launch
393 
394- This is normal - simply restart Claude Desktop
395- Wait 10-15 seconds before making requests
396 
397##### gget Commands Not Working
398 
399- Ensure internet connection is stable (gget fetches from online databases)
400- Check that the gget_mcp server shows as "Active" in settings
401- Try the verification command from Step 5 again
402 
403##### Google Drive Offline Access Issues
404 
405- Confirm the folder shows a green checkmark in Google Drive for Desktop
406- If files aren't syncing, right-click the folder and re-enable "Available offline"
407- Check that you have enough local disk space for offline files
408 
409#### Success! 🎉
410 
411If everything works correctly, you now have:
412- ✅ Claude Desktop with MCP capabilities
413- ✅ Access to biological databases via gget_mcp
414- ✅ Direct file operations in your Google Drive offline folder
415- ✅ Automatic cloud synchronization of all results
416 
417You can now ask Claude to fetch gene sequences, protein data, perform BLAST searches, and save everything directly to your Google Drive folder with offline access!
418 
419#### Example Use Cases
420 
421- **Fetch gene sequences:** "Get the DNA sequence for BRCA1 and save it to my Drive"
422- **Protein analysis:** "Find the protein sequence for TP53 and create a markdown report"
423- **Bulk operations:** "Search for all genes related to apoptosis and create a summary table"
424- **Cross-referencing:** "Get information about COL1A1 from Ensembl and UniProt"
425 
426#### Additional Resources
427 
428- [gget documentation](https://github.com/pachterlab/gget)
429- [Model Context Protocol docs](https://modelcontextprotocol.io/)
430- [Claude Desktop support](https://support.anthropic.com/)
431 
432*Guide updated: October 13, 2025*
433 
434</details>
435 
436## Repository setup
437 
438<details>
439<summary>For developers: cloning and running locally</summary>
440 
441```bash
442# Clone the repository
443git clone https://github.com/longevity-genie/gget-mcp.git
444cd gget-mcp
445uv sync
446```
447 
448### Running the MCP Server
449 
450If you already cloned the repo you can run the server with uv:
451 
452```bash
453# Start the MCP server locally (HTTP mode)
454uv run server
455 
456# Or start in STDIO mode  
457uv run stdio
458 
459# Or start in SSE mode
460uv run sse
461```
462 
463</details>
464 
465## Safety Features
466 
467- **Input validation**: Comprehensive parameter validation for all gget functions
468- **Error handling**: Robust error handling with informative messages
469- **Rate limiting**: Respectful usage of external APIs and databases
470- **Data validation**: Type checking and data format validation
471 
472## Testing & Verification
473 
474<details>
475<summary>Developer information: Testing and CI/CD workflows</summary>
476 
477The MCP server is provided with comprehensive tests including both unit tests and integration tests for network-dependent operations.
478 
479### Running Tests
480 
481Run tests for the MCP server:
482```bash
483# Run all tests (excluding expensive judge tests)
484uv run pytest -vvv -m "not judge"
485 
486# Run only fast tests (skip slow, integration, and judge tests)
487uv run pytest -vvv -m "not slow and not integration and not judge"
488 
489# Run only integration tests
490uv run pytest -vvv -m integration
491 
492# Run judge tests (expensive LLM tests - requires API keys and may cost money)
493uv run pytest test/test_judge.py -vvv
494 
495# Run tests with coverage (excluding judge tests)
496uv run pytest -vvv --cov=src/gget_mcp --cov-report=term-missing -m "not judge"
497```
498 
499**Note on Judge Tests**: The judge tests (`test_judge.py`) use large language models to evaluate the AI agent's performance with gget tools. These tests:
500- Are automatically excluded from CI/CD to avoid costs
501- Require API keys (GEMINI_API_KEY or similar) 
502- May incur charges from LLM API providers
503- Are designed for local development and manual evaluation
504- Provide valuable insights into how well the tools work with AI agents
505 
506### GitHub Actions Workflows
507 
508This project includes several GitHub Actions workflows:
509 
510- **CI** (`.github/workflows/ci.yml`): Runs on every push/PR with basic linting and fast tests
511- **Tests** (`.github/workflows/test.yml`): Comprehensive testing across multiple OS and Python versions
512  - Fast tests on all platforms
513  - Integration tests (may require network access)
514  - Slow tests (like BLAST operations) - runs only on Ubuntu with Python 3.11
515  - Code coverage reporting
516  - Security checks with bandit and safety
517- **Publish** (`.github/workflows/publish.yml`): Publishes to PyPI when tags are pushed
518 
519You can use MCP inspector with locally built MCP server same way as with uvx.
520 
521</details>
522 
523*Note: Using the MCP Inspector is optional. Most MCP clients (like Cursor, Windsurf, etc.) will automatically display the available tools from this server once configured. However, the Inspector can be useful for detailed testing and exploration.*
524 
525*If you choose to use the Inspector via `npx`, ensure you have Node.js and npm installed. Using [nvm](https://github.com/nvm-sh/nvm) (Node Version Manager) is recommended for managing Node.js versions.*
526 
527## Example Questions from Test Suite
528 
529Here are validated example questions that you can ask the AI assistant when using this MCP server:
530 
531* "Find information about the human TP53 gene and get its protein sequence."
532* "What are the orthologs of BRCA1 gene across different species?"
533* "Perform enrichment analysis for a set of cancer-related genes: TP53, BRCA1, BRCA2, ATM, CHEK2."
534* "Get the 3D structure information for the protein encoded by the EGFR gene."
535* "Find mutations in the COSMIC database for the PIK3CA gene."
536* "Analyze gene expression patterns for insulin (INS) gene across different tissues."
537* "Perform BLAST search with a DNA sequence to identify its origin: ATGGCGCCCGAACAGGGAC."
538* "Find diseases associated with the APOE gene using OpenTargets."
539* "Get reference genome information for mouse (Mus musculus)."
540* "Align multiple protein sequences and identify conserved regions."
541 
542These questions are taken directly from our automated test suite and are validated to work correctly with the available gget tools.
543 
544## Contributing
545 
546We welcome contributions from the community! 🎉 Whether you're a researcher, developer, or enthusiast interested in bioinformatics and genomics research, there are many ways to get involved:
547 
548**We especially encourage you to try our MCP server and share your feedback with us!** Your experience using the server, any issues you encounter, and suggestions for improvement are incredibly valuable for making this tool better for the entire research community.
549 
550### Ways to Contribute
551 
552- **🐛 Bug Reports**: Found an issue? Please open a GitHub issue with detailed information
553- **💡 Feature Requests**: Have ideas for new functionality? We'd love to hear them!
554- **📝 Documentation**: Help improve our documentation, examples, or tutorials
555- **🧪 Testing**: Add test cases, especially for edge cases or new query patterns
556- **🔍 Data Quality**: Help identify and report data inconsistencies or suggest improvements
557- **🚀 Performance**: Optimize queries, improve caching, or enhance server performance
558- **🌐 Integration**: Create examples for new MCP clients or AI systems
559- **🎥 Tutorials & Videos**: Create tutorials, video guides, or educational content showing how to use MCP servers
560- **📖 User Stories**: Share your research workflows and success stories using our MCP servers
561- **🤝 Community Outreach**: Help us evangelize MCP adoption in the bioinformatics community
562 
563**Tutorials, videos, and user stories are especially valuable to us!** We're working to push the bioinformatics community toward AI adoption, and real-world examples of how researchers use our MCP servers help demonstrate the practical benefits and encourage wider adoption.
564 
565### Getting Started
566 
5671. Fork the repository
5682. Create a feature branch (`git checkout -b feature/amazing-feature`)
5693. Make your changes and add tests
5704. Run the test suite (`uv run pytest`)
5715. Commit your changes (`git commit -m 'Add amazing feature'`)
5726. Push to your branch (`git push origin feature/amazing-feature`)
5737. Open a Pull Request
574 
575### Development Guidelines
576 
577- Follow the existing code style (we use `ruff` for formatting and linting)
578- Add tests for new functionality
579- Update documentation as needed
580- Keep commits focused and write clear commit messages
581 
582### Questions or Ideas?
583 
584Don't hesitate to open an issue for discussion! We're friendly and always happy to help newcomers get started. Your contributions help advance open science and bioinformatics research for everyone. 🧬✨
585 
586## Known Issues
587 
588<details>
589<summary>Technical limitations and considerations</summary>
590 
591### External Dependencies
592Some gget functions depend on external web services and databases that may occasionally be unavailable or rate-limited. The server implements proper error handling and retries where appropriate.
593 
594### Test Coverage
595While we provide comprehensive tests including integration tests for network-dependent operations, some test cases may be sensitive to external service availability. Some automated test results may need manual validation.
596 
597</details>
598 
599## License
600 
601This project is licensed under the MIT License.
602 
603## Acknowledgments
604 
605- [gget](https://github.com/pachterlab/gget) - The amazing bioinformatics toolkit this server wraps
606- [FastMCP](https://github.com/jlowin/fastmcp) - The MCP server framework used
607- [Model Context Protocol](https://modelcontextprotocol.io/) - The protocol specification
608 
609### Other MCP Servers by Longevity Genie
610 
611We also develop other specialized MCP servers for biomedical research:
612 
613- **[biothings-mcp](https://github.com/longevity-genie/biothings-mcp)** - MCP server for BioThings.io APIs, providing access to gene annotation (mygene.info), variant annotation (myvariant.info), and chemical compound data (mychem.info)
614- **[opengenes-mcp](https://github.com/longevity-genie/opengenes-mcp)** - MCP server for OpenGenes database, providing access to aging and longevity research data
615- **[synergy-age-mcp](https://github.com/longevity-genie/synergy-age-mcp)** - MCP server for SynergyAge database, providing access to synergistic and antagonistic longevity gene interactions
616 
617We are supported by:
618 
619[![HEALES](https://github.com/longevity-genie/biothings-mcp/raw/main/images/heales.jpg)](https://heales.org/)
620 
621*HEALES - Healthy Life Extension Society*
622 
623and
624 
625[![IBIMA](https://github.com/longevity-genie/biothings-mcp/raw/main/images/IBIMA.jpg)](https://ibima.med.uni-rostock.de/)
626 
627[IBIMA - Institute for Biostatistics and Informatics in Medicine and Ageing Research](https://ibima.med.uni-rostock.de/)
628