How to add citations to your AI chatbot

Send Tako's Answer API a question and get back a written answer, the data behind it and the sources it came from, in one call. Every answer carries its sources: licensed data providers and web pages, each named, so your chatbot can cite them, and show the data as a chart you can embed.

Get an API key Read the Answer docs

RESPONSE FIELDS

What each answer includes

Answer: A written answer to the question, grounded in the data and pages below.
Knowledge cards: The data behind the answer, with an embeddable chart. The first card is the one to show beside it.
Sources: Each card names its source and how the data was collected, for licensed providers like S&P Global and Semrush.
Web results: The pages the answer used, each with its title and URL.
Freshness: Each card in the response carries the date its data was last updated, and each web result its publish date, so your chatbot can show how current every source is.

QUICKSTART

Get a cited answer with one API call

Send the question as the query to the Answer API, with your API key in the X-API-Key header. The response is the answer as text, the knowledge cards it's built on, and the web results it used, so your chatbot can show the answer and its sources together. Every parameter is in the Answer API docs.

Python

import os
import requests

response = requests.post(
    "https://tako.com/api/v1/answer",
    headers={"X-API-Key": os.environ["TAKO_API_KEY"]},
    json={"query": "Did Apple beat analyst estimates last quarter, and by how much?"},
    timeout=60,
)
result = response.json()["outputs"]
print(result["answer"])
for card in result["knowledge_cards"]:
    print(card["title"], card["sources"][0]["source_name"])
for page in result["web_results"]:
    print(page["title"], page["url"])

RESPONSE

What the call returns

The answer comes back with everything it was built from: the answer text, the cards behind it, each naming its data provider, and the web pages it cited. Below is a real response, trimmed to the top card and the first web results.

The answer, its cards and its web results come back as JSON from the Answer API. Download a card's rows as CSV with the Contents API, or embed the card beside the answer.

In your chatbot

Did Apple beat analyst estimates last quarter, and by how much?

Yes. In fiscal Q3 2026, Apple’s EPS was $2.02 versus the $1.89 analyst estimate, a beat of $0.13 per share. ¹ ² ³

S&P Global apple.com apnews.com

JSON response

A real response, captured September 30, 2026.

{
  "outputs": {
    "answer": "Yes. In fiscal Q3 2026, Apple’s EPS was $2.02 versus the $1.89 analyst estimate, a beat of $0.13 per share.",
    "knowledge_cards": [
      {
        "card_id": "q2lARTdWHhZeipqtmQWV",
        "title": "Apple Earnings & Estimates Overview",
        "description": "Apple's latest EPS (Normalized), Q2 2026 (fiscal Q3 2026): consensus estimate $1.89 (range $1.82 to $1.99), actual $2.02, a beat of $0.13. …",
        "webpage_url": "https://tako.com/card/q2lARTdWHhZeipqtmQWV/",
        "embed_url": "https://tako.com/embed/q2lARTdWHhZeipqtmQWV/",
        "sources": [
          {
            "source_name": "S&P Global",
            "source_index": "data"
          }
        ],
        "card_type": "card",
        "exportable": false,
        "data_freshness": {
          "last_updated": "2026-09-28"
        }
      }
    ],
    "web_results": [
      {
        "title": "Apple reports third quarter results - Apple",
        "url": "https://www.apple.com/newsroom/2026/07/apple-reports-third-quarter-results/",
        "source_name": "apple.com",
        "publish_date": "2026-07-30T00:00:00.000Z"
      },
      {
        "title": "Apple's fiscal Q3 beats expectations with strong iPhone sales | AP News",
        "url": "https://apnews.com/article/apple-earnings-revenue-iphone-ai-94102918cb3592ebc1d2a38c4d7d819a",
        "source_name": "Associated Press",
        "publish_date": "2026-07-30T00:00:00.000Z"
      }
    ]
  },
  "request_id": "4b0d14b2-d0f5-4206-975b-316a5dc148e2"
}

Card

Need the numbers behind an answer, every quarter as a series? See how to give your AI agent company financial data.

CLAUDE, CURSOR, CHATGPT

Get cited answers in your AI assistant without code

Tako's MCP server gives any MCP client the same answers. No API key is needed: you sign in with a Tako account the first time you use it. Then ask in plain language, and the answer arrives with its sources.

Claude Code
Run claude mcp add --transport http tako https://mcp.tako.com/mcp, then /mcp to sign in.
Claude.ai
Open Settings, then Connectors, then Add custom connector, and paste https://mcp.tako.com/mcp.
Cursor
Add a server named tako with the URL https://mcp.tako.com/mcp to .cursor/mcp.json.
ChatGPT
In developer mode, add a plugin with the connector URL https://mcp.tako.com/mcp/chatgpt.

Every other client, from VS Code to Gemini CLI, is on the Tako MCP page.

FAQ

Frequently asked questions

What is a cited answer?

An answer that names where each fact came from. Tako's Answer API returns the answer with the data cards and web pages it was built from, so a reader can check each claim.

When should I use Answer instead of Search?

Use Answer when you want the written answer. Use Search when you want the data and web results to render or process yourself; it returns the same sources, without the answer.

Can I show citations inline in my chatbot?

Yes. Each response lists its cards and web results with their sources and URLs, so you can link them under the answer or beside the sentences they support.

How current is a cited answer?

As current as its sources. Each card in the response carries the date its data was last updated and each web result its publish date, so your chatbot can show readers how recent every source is.

Start building with Tako

Mistral's Le Chat and Liner embed Tako knowledge cards in their answers. Your first 2,000 requests are free.

Start free Read the docs