# Our models, as tools for your AI agent

Canonical: https://udgamlabs.com/mcp · Updated: 2026-10-31

MCP server · 11 tools

Add one address to Claude, Claude Code, Codex, Cursor or any app that speaks MCP (the Model Context Protocol), and nine Udgam Labs models become tools your agent can call: speech to text, Hindi transliteration, Telugu text, chat, maths, pictures, short video clips and on-device tool calling. Free, no key, no sign-up.

## The address

`https://mcp.udgamlabs.com/mcp`

Streamable HTTP, no sign-in. Opening [mcp.udgamlabs.com](https://mcp.udgamlabs.com) in a browser shows the same steps.

## Add it to your app

### Claude (web and desktop app)

1.  Open **Customize → Connectors** and choose **Add custom connector**.
2.  Paste the address above as the server URL. It needs no sign-in, so leave the authentication settings empty.
3.  Choose **Add**, then turn the connector on for a chat from the **+** menu.

Works on every Claude plan; the Free plan allows one custom connector. On Team and Enterprise plans, an Owner adds it under Organization settings → Connectors.

### Claude Code

In a terminal:

```
claude mcp add --transport http udgamlabs https://mcp.udgamlabs.com/mcp
```

### Codex

In a terminal:

```
codex mcp add udgamlabs --url https://mcp.udgamlabs.com/mcp
```

### Cursor

Add this to `.cursor/mcp.json` in your project, or to `~/.cursor/mcp.json` for every project:

```
{
  "mcpServers": {
    "udgamlabs": { "url": "https://mcp.udgamlabs.com/mcp" }
  }
}
```

### VS Code

Run **MCP: Add Server** from the Command Palette, choose **HTTP**, paste the address and name it `udgamlabs`.

### Any other MCP app

Add a remote (streamable HTTP) server with the address above. There is no key, and the server keeps no sessions.

## The tools

Your agent picks the tool from what you ask. Every result ends with a source line that names the model and links its page, where you can see its numbers and where it falls short.

- `shruti_transcribe` · [Shruti](https://udgamlabs.com/models/shruti) · Speech to text for Hindi, English and Hinglish, from a link to an audio file. · Runs on: Free CPU on Hugging Face
- `lekhani_transliterate` · [Lekhani](https://udgamlabs.com/models/lekhani) · Hindi typed in Latin letters to Devanagari (kya haal hai: क्या हाल है), with each word’s runners-up if asked. · Runs on: The MCP server itself
- `vemana_continue` · [Vemana](https://udgamlabs.com/models/vemana) · Continues Telugu text. A base model: it writes on, it does not answer questions. · Runs on: Free CPU on Hugging Face
- `laghu_chat` · [Laghu](https://udgamlabs.com/models/laghu) · One message to a 152M English chat model, with thinking on or off. · Runs on: Free CPU on Hugging Face
- `gargi_chat` · [Gargi](https://udgamlabs.com/models/gargi) · One message to a 384M English model that thinks step by step on maths problems. · Runs on: Free CPU on Hugging Face
- `lilavati_solve` · [Lilavati](https://udgamlabs.com/models/lilavati) · Solves a maths word problem by writing a short Python program and running it: the answer, how sure it is, and the program. · Runs on: Shared GPU on Hugging Face
- `chitra_tiny_generate_image` · [Chitra Tiny](https://udgamlabs.com/models/chitra-tiny) · A 256×256 picture from one sentence, returned as an image your agent can see. · Runs on: Shared GPU on Hugging Face
- `drishya_tiny_generate_video` · [Drishya Tiny](https://udgamlabs.com/models/drishya-tiny) · A 2-second, 256×256 video clip from a sentence, as a link. The weights are for non-commercial use. · Runs on: Shared GPU on Hugging Face
- `udgam_384m_tool_call` · [Udgam-384M](https://udgamlabs.com/models/udgam-384m) · Which tool call a 384M on-device model would make for a request, with your own tools. Nothing is run. · Runs on: Free CPU on Hugging Face
- `find_small_model` · Data · Measured scores of 27 open language models up to 400M parameters, from the Small Model Board. · Runs on: The MCP server itself
- `list_udgam_models` · Data · Every Udgam Labs model, with its size, licence, how to run it and where it falls short. · Runs on: The MCP server itself

## Things to ask

- Write “kal subah 10 baje milte hain” in Devanagari with Lekhani, and show me the words it was unsure about.
- Transcribe this recording with Shruti: <a direct link to a wav, mp3, flac or ogg file>
- Ask Lilavati: a phone costs 15,000 rupees, it is 20% off, then 5% tax is added to the sale price. What is the final price? Check its program.
- Draw a lighthouse on a rocky coast at sunset with Chitra Tiny, and tell me what you see.
- Which open language model under 200M parameters scores best? Use the Small Model Board.

## Good to know

**Speed.** Most models run on free shared CPUs on Hugging Face, so a call takes a few seconds, and the first call after a quiet spell waits while the model starts. Lekhani runs on the server itself, so a sentence takes about a second.

**Small models make mistakes.** Each tool tells your agent what the model is for and where it falls short; check anything that matters.

**Your data.** What your agent sends goes to the model only to produce the answer. We do not log inputs or answers, and we count calls without IP addresses ([privacy](https://udgamlabs.com/privacy)). To keep everything on your own device, run the models yourself: each [model page](https://udgamlabs.com/models) shows how.

**GPU models.** Lilavati, Chitra Tiny and Drishya Tiny run on Hugging Face’s shared GPUs (ZeroGPU), which give callers a small free allowance each day. If it runs out, add your own Hugging Face token as the header `X-HF-Token` to use your account’s daily GPU minutes. We pass it to Hugging Face for that call only and never store it.

```
# Claude Code
claude mcp add --transport http udgamlabs https://mcp.udgamlabs.com/mcp --header "X-HF-Token: hf_..."

# Cursor: .cursor/mcp.json
{ "mcpServers": { "udgamlabs": { "url": "https://mcp.udgamlabs.com/mcp", "headers": { "X-HF-Token": "${env:HF_TOKEN}" } } } }

# Codex: ~/.codex/config.toml (the token is read from the HF_TOKEN environment variable)
[mcp_servers.udgamlabs]
url = "https://mcp.udgamlabs.com/mcp"
env_http_headers = { "X-HF-Token" = "HF_TOKEN" }
```

---

## About Udgam Labs

Udgam Labs is an independent AI research lab. We work from first principles: before we build, we understand why things work, whether we are training a model, fine-tuning one or applying AI to a new problem. Today that means open models such as Gargi, a 384M-parameter language model that comes within 2 points of Qwen2.5-0.5B on three reasoning benchmarks with about 1,000× less training data. Next come products, custom models for businesses, and AI in the physical world. Udgam (उद्गम) is Sanskrit for origin.

Udgam Labs · AI research from first principles · https://udgamlabs.com