# grad-agent [Health: Active]

**Category:** 🎓 Education  
**Repository:** https://github.com/i-ninte/grad-agent  
**GitHub Stars:** 4  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/grad-agent

## Description
Autonomous MCP agent for grad school applications: prof discovery, cold emails, SOP, tracking.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "grad-agent": {
    "command": "uvx",
    "args": ["grad-agent"],
    "env": {
      "ANTHROPIC_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `ANTHROPIC_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What grad-agent MCP server does

The grad-agent MCP server supports graduate-school outreach and application tracking for applicants targeting fully funded MS or PhD programs. It can discover professors through arXiv research areas or program deadlines, resolve identities through Semantic Scholar author IDs, and reject ambiguous matches instead of relying on name strings alone. Faculty checks include h-index and publication-count gates, while region filtering can use profile settings, keyword data, homepage domains, and an LLM fallback.

For selected faculty, it examines lab pages for recruitment signals and contact details, then compares their research with the applicant’s projects using tags, TF-IDF similarity, and learned response outcomes. Cold-email drafts reference recent papers, receive a fit score from 1 to 10, and undergo a second verification pass against paper abstracts. Drafts can include warnings for stale publication activity or disagreements between Semantic Scholar affiliation data and a live homepage.

The system also detects replies through read-only IMAP, prepares follow-up drafts for contacts silent for at least 10 days, and creates interview preparation documents after replies. It compiles school-specific SOPs to versioned LaTeX PDFs and tracks outreach, recommendation-letter requests, deadlines, and external scholarships in YAML or XLSX files.

## How it works

The grad-agent MCP server reads configuration from `~/.grad-agent`, including the applicant profile, target programs, scholarships, and secrets. Profile changes apply while a long-running MCP session remains open. A project synchronization command can scan GitHub, Hugging Face, and local projects before a batch run produces drafts for review.

Daily execution can be scheduled with launchd, a systemd user timer, or Windows Task Scheduler. Bundled Claude skills cover SOP writing and recommendation-letter writing, and the package also includes a cold-email drafter.

## Setup and configuration

Install with `uvx grad-agent server`, `pipx install grad-agent`, or a virtual-environment installation. Run `grad-agent init` to create the profile, program, scholarship, and environment files. Configure the Anthropic key, SMTP credentials, and the applicant’s identity, documents, degree status, target term, degree, regions, research areas, and seed projects. Semantic Scholar, IMAP, GitHub, and Hugging Face settings are available for their respective features.

Claude Code and Claude Desktop can launch the server with `grad-agent server` or `uvx grad-agent server`. Cursor, Zed, and Windsurf use the same stdio manifest pattern. The package documentation also provides a command for installing its Claude skills and an optional letterhead configuration for recommendation-letter PDFs.

## Tools and capabilities

- Discover and batch-process potential faculty leads.
- Verify identities and record skipped ambiguous matches.
- Filter by region, faculty indicators, recruitment signals, and research fit.
- Draft, fact-check, score, and email outreach for review.
- Detect replies and prepare one-time follow-up nudges.
- Generate interview preparation sheets and versioned SOP PDFs.
- Track programs, scholarships, outreach, and recommendation requests.

## Limitations and notes

The server depends on correctly completed profile and environment files. Identity resolution can skip candidates when comparable Semantic Scholar matches cannot be distinguished. IMAP reply detection is read-only, and outreach remains a review workflow rather than automatic professor delivery. API rate limits and the quality of public faculty pages or paper abstracts affect discovery and verification results.

_Full upstream README: https://allmcps.com/mcp/grad-agent/readme_