Skip to content
Tanvrit AI
Web App
Integrations

Providers and clients that work with Tanvrit AI

Two sides of the same coin: the 10 AI providers Tanvrit AI routes your agent's LLM calls to, and the MCP clients that talk to its local tool server on localhost:19281.

Browse AI providersConnect an MCP client

Cloud models are bring-your-own-key. Local inference and semantic code search run fully offline, on-device.

AI providers
10

8 live · 2 beta, behind one router

MCP tools
45

one bearer token, one URL

Local MCP port
19281

Authorization: Bearer

AI providers

Ten providers behind one router

Bring your own API keys in Settings → Providers. The agent picks one per request via the priority list, and auto-falls-back on network or rate-limit errors. Status reflects what actually ships in v1.5.29 — verified against the providers registered in AIProviderManager.

8 live · 2 beta · 2 planned

Tanvrit Local

tanvrit-local

Live
LocalIn-process local chat

llama.cpp via JNI, no daemon or Docker. Default landing provider on a fresh install; falls through automatically on platforms without a real binding.

Anthropic Claude

claude

Live
CloudChat, tools, streaming

Sonnet, Opus, Haiku families. Bring your own ANTHROPIC_API_KEY in Settings → Providers.

OpenAI

openai

Live
CloudChat, tools, streaming

GPT-4o, GPT-4.1, o-series. Tanvrit also ships an OpenAI-compatible local API on port 19280.

Google Gemini

gemini

Live
CloudChat, tools, streaming

Gemini 1.5 / 2.x via Generative Language API key.

DeepSeek

deepseek

Live
CloudChat, streaming

DeepSeek Chat and Coder. OpenAI-compatible endpoint.

Groq

groq

Live
CloudChat, streaming

Llama 3 / Mixtral hosted on Groq's LPU stack. OpenAI-compatible.

Mistral AI

mistral

Live
CloudChat, tools, streaming

Mistral Large / Codestral via la Plateforme API.

Ollama

ollama

Live
LocalLocal chat, streaming

Auto-detected on localhost:11434. Recommended pulls: qwen2.5-coder, nomic-embed-text, qwen2.5-vl. See /features/local-inference.

Local ML (Core ML)

local-ml

Beta
LocalLocal chat via Swift engine

macOS/Linux (Core ML backend on macOS). Launched as a child process. Wasm portal does not ship this provider; engine-dependent features degrade silently in the browser.

Tanvrit Mesh

mesh

Beta
LocalChat (text-only; no tool calls yet)

Routes a chat turn to a peer/fleet node over the mesh control plane instead of a cloud LLM. Falls back to a cloud provider on no eligible node — except on the Enterprise tier, where nothing leaves the org.

Azure OpenAI

azure-openai

Planned
Cloud—

Tracked for a future minor release. Until then, point an OpenAI-compatible config at your Azure endpoint via the OpenAI provider's base-URL override.

AWS Bedrock

bedrock

Planned
Cloud—

Tracked. Bedrock-native auth (SigV4) is not yet shipped.

Want a provider we don't list? Let us know via Support. Most OpenAI-compatible endpoints can be wired today by overriding the base URL on the OpenAI provider.

Domain Packs

Beyond LLM providers, Domain Packs integrate Tanvrit's business platforms with the agent runtime — governed, approval-gated, and off by default.

Commerce pack v1 — live

The Commerce pack v1 ships 12 governed commerce_* tools that ride the existing /api/ai/v1 gateway, authenticated with an AiScope-scoped X-API-Key minted on the Tanvrit server and stored in your OS keychain. Enable packs individually in Settings → Domain Packs (default off); pack tools register on both the in-app agent and the MCP tools/list — on top of the 45 built-in tools — and every mutating call is approval-gated.

How packs and agents are governed

MCP clients

Same token, same URL — only the config changes

Tanvrit AI's MCP server is at http://localhost:19281 with Authorization: Bearer auth. Stdio mode is also supported for trusted parents. Drop the matching block into your client and restart.

Claude Desktop

Edit the Claude Desktop config and restart. The MCP plug icon should turn green within a few seconds.

// macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
// Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "tanvrit-ai": {
      "type": "http",
      "url": "http://localhost:19281",
      "headers": {
        "Authorization": "Bearer YOUR_TANVRIT_BEARER_TOKEN"
      }
    }
  }
}

Cursor

One-click coming

Drop this block into Cursor’s MCP settings. When Cursor ships a one-click MCP add UI we will publish a deep link from this page.

// Cursor → Settings → MCP
{
  "mcp.servers": {
    "tanvrit-ai": {
      "transport": "http",
      "url": "http://localhost:19281",
      "headers": { "Authorization": "Bearer YOUR_TANVRIT_BEARER_TOKEN" }
    }
  }
}

Zed

Zed’s context-server contract uses a command + args. The curl invocation below streams Tanvrit AI’s SSE channel.

// Zed → settings.json
{
  "context_servers": {
    "tanvrit-ai": {
      "command": {
        "path": "curl",
        "args": ["-N",
                 "-H", "Authorization: Bearer YOUR_TANVRIT_BEARER_TOKEN",
                 "http://localhost:19281"]
      }
    }
  }
}

Continue.dev

Add Tanvrit AI to ~/.continue/config.json as an MCP server. Continue will surface tools in its tool picker on next launch.

// ~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "tanvrit-ai",
      "transport": "http",
      "url": "http://localhost:19281",
      "headers": { "Authorization": "Bearer YOUR_TANVRIT_BEARER_TOKEN" }
    }
  ]
}

Editors & IDEs

Built-in editor surfaces

The VS Code extension self-manages its own runtime and talks to the same local MCP server — no separate desktop-app install needed.

v1.5.1 — live

VS Code extension

Install the Tanvrit AI extension from the VS Code Marketplace. v1.5.1 makes it a full copilot surface: streaming agent runs in a panel, Pending Edits review (apply or reject each staged hunk from the editor — the extension never writes files itself), editor context sync, and agent run history. It talks to the same local MCP server at http://localhost:19281/ and pairs fully with desktop v1.5.29. Note: extensions 1.3.x and older can no longer run agents against desktop 1.5.29 — agent endpoints are bearer-gated now; upgrade to 1.5.1.

Open MarketplacePlatform overview
Phase 2

JetBrains (IntelliJ, GoLand, Rider…)

On the roadmap — until shipped, JetBrains users can run the desktop app alongside their IDE and use any MCP-aware client (e.g. the in-built AI chat once JB lands MCP).

Build your own

Anything that speaks JSON-RPC works

Wire a custom agent straight to the MCP server, then grab the bearer token from the desktop app to authenticate every client above.

Custom HTTP MCP agent

Anything that speaks JSON-RPC works. Here's the minimal request shape:

// Custom agent — minimal HTTP MCP request
POST http://localhost:19281
Authorization: Bearer YOUR_TANVRIT_BEARER_TOKEN
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list"
}

See the full 45-tool reference for the whole vocabulary.

Get the bearer token

Open the Tanvrit AI desktop app → Settings → MCP Bearer Token. The token is generated on first launch and stored in your platform keychain. Rotate it any time — the old token is revoked immediately.