Agents: use Tapetide via MCP or REST

Tapetide has two front doors to the same data, account and quota.

	MCP server	REST API
Endpoint	`https://mcp.tapetide.com/mcp` (Streamable HTTP)	`https://api.tapetide.com/v1`
Best for	Chat apps and coding assistants that speak MCP	Your own code, pipelines and custom agents
Auth	Google sign-in (OAuth) or a personal token	Personal token
Shape	55 tools with LLM-oriented descriptions	Typed JSON documents described by OpenAPI 3.1

Calls on either side count against one quota: 50 successful calls a day on the free plan.

Connect an MCP client

Most clients only need the server URL, then a Google sign-in when the browser opens.

Claude Code

claude mcp add --transport http tapetide https://mcp.tapetide.com/mcp

Then open /mcp in Claude Code, select Tapetide and sign in.

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "tapetide": { "url": "https://mcp.tapetide.com/mcp" }
  }
}

Codex

codex mcp add tapetide --url https://mcp.tapetide.com/mcp
codex mcp login tapetide

Claude.ai and ChatGPT: add a custom connector named Tapetide with the URL https://mcp.tapetide.com/mcp and sign in with Google.

Clients that cannot open a browser send a personal token instead: Authorization: Bearer tpt_rt_.... Claude Desktop uses the local helper:

{
  "mcpServers": {
    "tapetide": {
      "command": "bunx",
      "args": ["tapetide-mcp"],
      "env": { "TAPETIDE_TOKEN": "YOUR_TOKEN" }
    }
  }
}

Step-by-step pages for each client are canonical on tapetide.com:

Chat apps: Claude.ai, ChatGPT, Grok, Claude Desktop
Editors: Cursor, VS Code, Windsurf, Zed, Kiro, Antigravity
Agents and CLIs: Claude Code, Codex, Gemini CLI, Cline, OpenCode, Command Code

Confirm the connection with a prompt such as “Use Tapetide to show Reliance’s latest price and when it was updated.” The answer should include a timestamp.

Full MCP tool reference: tapetide.com/mcp/llms-full.txt. For when to pick which, see MCP vs API for Indian stock data.

Give your own agent the REST API

If you are building the agent, REST is usually simpler to control and test:

Load the OpenAPI document from https://api.tapetide.com/v1/openapi.json and turn operations into tools. Most agent frameworks accept OpenAPI directly.
Point the model at /llms.txt or /llms-full.txt for the conventions: units, nulls, paging.
Put the token in the HTTP client, never in the prompt.
Have the agent quote meta.as_of and any observation timestamp (updated_at, snapshot_ts) with numbers it reports.

A minimal tool for an agent loop:

def tapetide_get(path: str, **params) -> dict:
    """GET a Tapetide API path such as /v1/stocks/TCS/quote. Returns the JSON body."""
    r = session.get(f"https://api.tapetide.com{path}", params=params, timeout=30)
    return r.json()  # errors come back as {"error": {"code", "message"}} for the model to read

Returning error bodies to the model works well: messages name the bad parameter and list accepted values.