MCP Server - Tako Docs

Tako runs a hosted, remote Model Context Protocol server, so any MCP-compatible client — Claude, ChatGPT, Cursor, and others — can search, answer, and fetch data from Tako without you writing any glue code. The model decides when to call Tako; the server runs the call and returns grounded, up-to-date results plus the interactive charts behind them. There is nothing to install or host: the server is a single endpoint.

https://mcp.tako.com/mcp

It speaks Streamable HTTP (the current MCP remote transport).

This page covers the three data tools — tako_search, tako_answer, and tako_contents — which mirror Tako’s Search, Answer, and Contents APIs, plus tako_visualize (build a chart from your own data — Tako’s Thin-Viz API), the deep-research tako_agent, and the get_credit_balance utility.

Connect

Your client decides how it authenticates — you don’t have to choose. Web hosts (Claude.ai, ChatGPT, Claude Desktop) open a Tako sign-in window the first time you add the connector; approve once and nothing goes in a config file. CLI and editor clients send a Bearer token in an Authorization: Bearer tako_sk_… header — though most of them can do the sign-in flow instead if you leave the header out. Mint a token at tako.com/dashboard → Settings → API tokens. Tokens look like tako_sk_…, are shown once, and each connected client appears as its own named, revocable key. The endpoint is the same everywhere: https://mcp.tako.com/mcp.

Claude.ai

Requires a Claude Pro, Max, Team, or Enterprise plan.

Open Settings → Connectors → Add custom connector.
Enter https://mcp.tako.com/mcp and click Add.
Approve the Tako sign-in window when it opens — the Tako tools then appear. No token required.

On Team/Enterprise, a workspace owner adds the connector; members then click Connect to sign in individually.

ChatGPT

Requires ChatGPT Plus, Pro, Business, Enterprise, or Edu.

Web only. The connector setup lives in the ChatGPT web app — Developer mode and custom connectors aren’t available in the desktop or mobile apps.

Open Settings → Apps (formerly “Connectors”) → Advanced settings → Developer mode and toggle it on.
Click Create app, give it a name, and set the Connector URL to https://mcp.tako.com/mcp.
Approve the Tako sign-in when prompted. No token required.

Claude Code

Add Tako with the CLI, passing a Bearer token:

claude mcp add --transport http tako https://mcp.tako.com/mcp \
  --header "Authorization: Bearer tako_sk_YOUR_TOKEN"

Or connect via OAuth — omit the header, then run /mcp inside Claude Code and approve the Tako sign-in:

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

Verify with claude mcp list.

Claude Desktop

Open Settings → Connectors → Add custom connector.
Enter https://mcp.tako.com/mcp.
Sign in when the Tako window opens.

Claude Desktop connects through Anthropic’s cloud and does not accept a static Bearer token — use the OAuth sign-in here. (To run Tako locally with a token instead, see Other.)

Cursor

The button installs Tako and starts the sign-in flow — no token needed. Or add it to ~/.cursor/mcp.json (or a project .cursor/mcp.json) by hand:

{
  "mcpServers": {
    "tako": {
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:TAKO_API_KEY}"
      }
    }
  }
}

Cursor auto-detects the transport. To sign in with OAuth instead of a token, omit headers.

VS Code

The button installs Tako and starts the sign-in flow. Or add it to .vscode/mcp.json by hand:

{
  "servers": {
    "tako": {
      "type": "http",
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer ${input:tako-key}"
      }
    }
  },
  "inputs": [
    { "id": "tako-key", "type": "promptString", "description": "Tako API key", "password": true }
  ]
}

The top-level key is servers (in user settings.json it nests under "mcp"). Omit headers and inputs to sign in with OAuth instead.

Windsurf

Add Tako to ~/.codeium/windsurf/mcp_config.json — note the key is serverUrl, not url:

{
  "mcpServers": {
    "tako": {
      "serverUrl": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:TAKO_API_KEY}"
      }
    }
  }
}

Codex

Use the Codex CLI; it reads the token from an environment variable rather than storing it in the config file:

export TAKO_API_TOKEN="tako_sk_YOUR_TOKEN"
codex mcp add tako --url https://mcp.tako.com/mcp --bearer-token-env-var TAKO_API_TOKEN

That writes the following to ~/.codex/config.toml, which you can also add by hand:

[mcp_servers.tako]
url = "https://mcp.tako.com/mcp"
bearer_token_env_var = "TAKO_API_TOKEN"

Or sign in via OAuth with codex mcp login tako.

Gemini CLI

Add Tako with the CLI — the Streamable HTTP key is httpUrl:

gemini mcp add --transport http \
  --header "Authorization: Bearer tako_sk_YOUR_TOKEN" \
  tako https://mcp.tako.com/mcp

Equivalent ~/.gemini/settings.json:

{
  "mcpServers": {
    "tako": {
      "httpUrl": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Zed

Add a context server to Zed’s settings.json (Zed calls MCP servers “context servers”):

{
  "context_servers": {
    "tako": {
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Omit headers to sign in with OAuth instead.

Cline

In the Cline MCP Servers panel, open the Remote Servers tab, set the URL to https://mcp.tako.com/mcp, and choose the Streamable HTTP transport. The equivalent JSON config:

{
  "mcpServers": {
    "tako": {
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Roo Code

Open Roo Code’s MCP settings (Edit Global MCP) and add Tako — the type must be streamable-http:

{
  "mcpServers": {
    "tako": {
      "type": "streamable-http",
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Kiro

The button installs Tako and starts the sign-in flow. Or add it to ~/.kiro/settings/mcp.json (or workspace .kiro/settings/mcp.json) by hand:

{
  "mcpServers": {
    "tako": {
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Omit headers to sign in with OAuth instead.

Warp

In Warp, open Settings → Agents → MCP servers → + Add, choose the URL (Streamable HTTP) option, and paste:

{
  "mcpServers": {
    "tako": {
      "url": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Warp also offers a one-click OAuth sign-in if you add the server by URL without a header.

Goose

Run goose configure → Add Extension → Remote Extension (Streamable HTTP), set the endpoint to https://mcp.tako.com/mcp, then add a Custom Request Header: Authorization: Bearer tako_sk_YOUR_TOKEN. The Goose desktop app has the same flow under Extensions → Add custom extension.

Antigravity

In Antigravity, open the side-panel ”…” menu → Manage MCP Servers → View raw config (the raw-config view shows the exact file path for your install), and add Tako — note the key is serverUrl:

{
  "mcpServers": {
    "tako": {
      "serverUrl": "https://mcp.tako.com/mcp",
      "headers": {
        "Authorization": "Bearer tako_sk_YOUR_TOKEN"
      }
    }
  }
}

Omit headers to sign in with OAuth instead.

Other

Any client that speaks Streamable HTTP can use Tako: point it at https://mcp.tako.com/mcp and either complete the OAuth sign-in or send Authorization: Bearer tako_sk_…. For clients that only support stdio servers, bridge to the remote endpoint with mcp-remote:

{
  "mcpServers": {
    "tako": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.tako.com/mcp", "--header", "Authorization:${AUTH_HEADER}"],
      "env": { "AUTH_HEADER": "Bearer tako_sk_YOUR_TOKEN" }
    }
  }
}

The header is supplied through an env var with no space after the colon — the documented workaround for an npx argument-escaping bug on some clients. To use OAuth instead, drop --header/env and run the bare bridge: npx mcp-remote https://mcp.tako.com/mcp.

Tools

`tako_search`

Find live data for a specific, known thing and get back ranked knowledge cards — each a structured factual answer with an interactive chart — backed by Tako’s curated knowledge graph and the live web. On hosts that support it, the top card renders inline automatically; reference it in your reply and link embed_url so the user can open it fullscreen. Use it for a current or latest value, a time series, a statistic, a price, a score, a schedule, a forecast, a poll, or prediction-market odds — including a direct comparison of two named entities — across sports, economics, finance, demographics, technology, weather, elections, prediction markets, web traffic, real estate, energy, and health. When the question instead needs figuring something out — resolving a cohort, ranking or filtering a set by criteria, or multi-step reasoning across many entities — use tako_agent instead.

Parameter	Type	Default	Description
`query`	string	— (required)	Natural-language query, e.g. `"US GDP growth"` or `"Intel vs Nvidia revenue"`.
`count`	integer (1–20)	`10`	Maximum number of cards to return.
`sources`	array of `"data"` \| `"web"`	`["data","web"]`	Where to search. Defaults to both Tako’s curated cards and the live web; pass `["data"]` for curated cards only or `["web"]` for web results only. The legacy value `tako` is accepted as a synonym for `data`.
`effort`	`"instant"` \| `"fast"`	(auto)	`instant` is the fastest cached path; `fast` runs the standard synchronous search. Omit to let Tako choose.
`include_contents`	boolean	`false`	Inline each result’s underlying data in the response (a `content` object): the Tako card’s CSV and any web page’s text, fetched in the same call. Leave off and use `tako_contents` to fetch data on demand.
`country_code`	string	`"US"`	ISO country code for localized results.
`locale`	string	`"en-US"`	Locale for results.

Returns cards (each with card_id, title, description, webpage_url, image_url, embed_url, and — when include_contents is set — an inline content object), web_results (title, url, snippet, source_name, with inline content when requested; web results are not chartable cards), contents_total_cost (USD for any inlined data, 0 otherwise), and request_id. When the top card is chartable, the response also lifts its embed_url and image_url to the root for inline rendering.

Example

Prompt: “How has Nvidia’s revenue grown since 2015?”

// tako_search({ "query": "Nvidia revenue since 2015" })
{
  "cards": [
    {
      "card_id": "j60lxSe9mhx4-ni0ylTk",
      "title": "NVIDIA Corporation Revenue",
      "description": "Annual revenue, FY2015–FY2024…",
      "webpage_url": "https://tako.com/card/j60lxSe9mhx4-ni0ylTk",
      "embed_url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk",
      "image_url": "https://tako.com/api/v1/image/j60lxSe9mhx4-ni0ylTk/"
    }
  ],
  "web_results": [],
  "contents_total_cost": 0,
  "request_id": "req_abc123",
  "embed_url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk",
  "image_url": "https://tako.com/api/v1/image/j60lxSe9mhx4-ni0ylTk/"
}

`tako_answer`

Ask a factual question about a specific, known thing and get back a single grounded, citation-backed text answer (prose, not a chart), plus the backing cards and web results. Use it when the user wants a direct written answer rather than a rendered chart. When the question needs figuring something out — cohort resolution, ranking or filtering a set, multi-step reasoning across many entities — use tako_agent instead.

Parameter	Type	Default	Description
`query`	string	— (required)	Natural-language question, e.g. `"What was US GDP in 2024?"`.
`sources`	array of `"data"` \| `"web"`	`["data","web"]`	Where to ground the answer. Defaults to both Tako’s curated data and the live web; pass `["data"]` for curated data only or `["web"]` for live web only. The legacy value `tako` is accepted as a synonym for `data`.
`include_contents`	boolean	`false`	Inline the underlying data of each backing result (`content`): the card’s CSV and any web page’s text. Leave off and use `tako_contents` to fetch data on demand.
`country_code`	string	`"US"`	ISO country code for localized results.
`locale`	string	`"en-US"`	Locale for results.

Returns answer (the synthesized text), cards (backing knowledge cards with card_id, title, description, webpage_url, image_url, embed_url, plus inline content when include_contents is set), web_results (title, url, snippet, source_name), contents_total_cost (USD for any inlined data, 0 otherwise), and a request_id.

Example

Prompt: “What is the current US unemployment rate?”

// tako_answer({ "query": "current US unemployment rate" })
{
  "answer": "The current US unemployment rate is 4.1% (latest monthly reading).",
  "cards": [
    {
      "card_id": "j60lxSe9mhx4-ni0ylTk",
      "title": "United States Unemployment Rate",
      "embed_url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk",
      "webpage_url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk"
    }
  ],
  "web_results": [],
  "contents_total_cost": 0,
  "request_id": "req_abc123"
}

tako_search and tako_answer take nearly the same inputs and differ only in what comes back: tako_search returns cards to render, tako_answer returns prose plus the backing cards. Reach for tako_answer when you want a written answer read back to the user, and tako_search when you want to display or post-process the charts.

`tako_agent`

Tako’s deep research agent — for questions that require figuring something out rather than retrieving a known value: resolving a cohort (“which companies match…”), ranking or filtering a set by criteria, multi-step aggregation, or multi-hop reasoning across many entities that a single tako_search / tako_answer can’t satisfy. It researches across Tako and the web and returns a synthesized answer plus the supporting chart cards and web_results. Runs server-side (typically ~30–90s).

Parameter	Type	Default	Description
`query`	string	— (required)	The analytical question to work through, e.g. `"Which of the largest US semiconductor companies grew revenue fastest since 2015?"`.
`sources`	array of `"data"` \| `"web"`	`["data","web"]`	Where the agent researches. Defaults to both; pass `["data"]` for curated data only or `["web"]` for the live web only. The legacy value `tako` is accepted as a synonym for `data`.
`thread_id`	string	—	A `thread_id` from a prior run, to ask a follow-up in the same conversation. Omit to start a new thread.

Returns an agent run: run_id, thread_id, status (queued → running → completed / failed), and on completion a result carrying the synthesized answer, the backing cards, web_results, and a request_id.

On ChatGPT the agent is split into two tools — tako_agent_start (dispatch; returns a run_id) and tako_agent_wait (poll until done) — because the single-call path can’t survive that host’s tool-call timeout. Every other host exposes a single tako_agent tool that dispatches and waits in one call.

See the Agent guide for the dispatch / poll / stream flow over the HTTP API.

`tako_contents`

Fetch the content behind a result you already have, given that result’s URL. A Tako card URL resolves to a CSV of the card’s underlying dataset; any other URL resolves to the page’s extracted full text. One URL per call. Use it after a tako_search / tako_answer result when the user wants the raw data or text — the input is a result URL, not a search query. mode controls delivery: inline (the default) returns the data in the response so the model can read it; url returns a presigned download link instead.

Parameter	Type	Default	Description
`url`	string	— (required)	The result URL to fetch — a Tako card’s `webpage_url`/`embed_url` (→ CSV) or a web result’s `url` (→ page text).
`mode`	`"inline"` \| `"url"`	`"inline"`	`inline` returns the content in the response (`data`) so the model can read it — CSV is capped at 1000 rows (`total_rows` / `truncated` flag the cap); `url` returns a short-lived `download_url` (no row cap), for handing the user a file or for large datasets you don’t need to read.

Returns format (e.g. csv or text), the source_url, and cost (USD for the fetch — Tako-card CSV is free; web text is metered). In inline mode it also returns data (the content, with total_rows / truncated describing CSV truncation) and leaves download_url / expires_at null. In url mode it returns a short-lived download_url + expires_at and leaves data null.

Example

Prompt: “Get the data behind that Nvidia chart.”

// tako_contents({ "url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk" })  // mode defaults to "inline"
{
  "format": "csv",
  "data": "fiscal_year,revenue_usd_b\n2015,5.0\n2016,6.9\n…\n2024,60.9",
  "total_rows": 10,
  "truncated": false,
  "download_url": null,
  "expires_at": null,
  "source_url": "https://tako.com/embed/j60lxSe9mhx4-ni0ylTk",
  "cost": 0
}

`tako_visualize`

Build a brand-new interactive Tako chart from your own data — the Thin-Viz chart-creation tool (it posts to Tako’s Thin-Viz create endpoint). This is the only write tool on the server: it mints a publicly hostable chart and returns its URLs. Pass an array of component configs (optionally a header, plus one or more visualization components); Tako renders, hosts, and returns the card. Auto-renders inline when the host supports it.

Parameter	Type	Default	Description
`components`	array of objects	— (required)	One or more component configs. Each has a `component_type` (e.g. `"header"`, `"generic_timeseries"`, `"categorical_bar"`, `"choropleth"`, `"pie"`, `"scatter"`, `"treemap"` — 15+ types), an optional `component_variant`, and a `config` object whose shape depends on the component type.
`title`	string	header title	Card title. Falls back to the header component’s title if omitted.
`description`	string	—	Optional card description.
`source`	string	—	Optional data-source attribution shown in the card footer (e.g. `"Yahoo Finance"`, `"Company Reports"`).
`height`	integer	—	Optional chart height in pixels (100–2000); overrides the default aspect-ratio height.
`normalize_currencies`	string	—	Optional target ISO 4217 currency code (e.g. `"USD"`) to convert currency-denominated datasets.

Returns card_id, title, description, webpage_url, embed_url, and image_url for the new card. On hosts that support MCP Apps the card auto-renders inline; otherwise share the returned embed_url (or webpage_url). See Visualize Your Data for an agent-ready setup.

Example

Prompt: “Chart sales by region: North America 500, Europe 300 (USD).”

// tako_visualize({
//   "title": "Sales by Region",
//   "components": [
//     { "component_type": "categorical_bar",
//       "config": { "datasets": [{ "label": "Sales", "units": "USD",
//         "data": [{ "x": "North America", "y": 500 }, { "x": "Europe", "y": 300 }] }] } }
//   ]
// })
{
  "card_id": "k71pYf3aZ2qb-rs9wmUv",
  "title": "Sales by Region",
  "description": null,
  "webpage_url": "https://tako.com/charts/k71pYf3aZ2qb-rs9wmUv",
  "embed_url": "https://tako.com/embed/k71pYf3aZ2qb-rs9wmUv",
  "image_url": "https://tako.com/api/v1/image/k71pYf3aZ2qb-rs9wmUv/"
}

`get_credit_balance`

Return the current account’s remaining Tako credits (plus subscription metadata). Call it only when the user explicitly asks about their balance, credits, or subscription — Tako’s other tools surface their own cost in each response, so a pre-flight balance check before another call is redundant and noisy. Takes no parameters. Returns details, an object carrying credit_balance (the remaining balance, a number) and formatted_credit_balance (a display string), plus any additional subscription / usage fields.

Example

Prompt: “How many Tako credits do I have left?”

// get_credit_balance({})
{
  "details": {
    "credit_balance": 4.85,
    "formatted_credit_balance": "$4.85"
  }
}

Troubleshooting

Sign-in required / no API token

Tools don't appear after connecting

Confirm your client supports the Streamable HTTP transport and is pointed at https://mcp.tako.com/mcp (not a .../sse path). For header-based clients, make sure the Authorization: Bearer tako_sk_… header is set and the token hasn’t been revoked.

​Connect

​Claude.ai

​ChatGPT

​Claude Code

​Claude Desktop

​Cursor

​VS Code

​Windsurf

​Codex

​Gemini CLI

​Zed

​Cline

​Roo Code

​Kiro

​Warp

​Goose

​Antigravity

​Other

​Tools

​tako_search

​tako_answer

​tako_agent

​tako_contents

​tako_visualize

​get_credit_balance

​Troubleshooting

Connect

Claude.ai

ChatGPT

Claude Code

Claude Desktop

Cursor

VS Code

Windsurf

Codex

Gemini CLI

Zed

Cline

Roo Code

Kiro

Warp

Goose

Antigravity

Other

Tools

`tako_search`

`tako_answer`

`tako_agent`

`tako_contents`

`tako_visualize`

`get_credit_balance`

Troubleshooting