Connect the Mantic MCP server
Let Claude, ChatGPT, Codex and other MCP clients search and explain Mantic forecasts.
The Mantic MCP (Model Context Protocol) server lets an AI assistant search questions, read the latest forecast and its reasoning, and follow how a forecast has changed over time.
Connection details
| Server URL | https://api.mantic.com/mcp |
| Transport | Streamable HTTP |
| Authentication | OAuth: sign in with your Mantic account |
You need a Mantic account. Through MCP you see the same questions and dashboards your account sees at app.mantic.com.
Connect a client
Claude (claude.ai and Claude Desktop)
Install the Mantic plugin, which adds the MCP server and skills that help Claude find and explain forecasts:
- Open Customize → Plugins.
- Choose Add → Add marketplace and enter
mantic-ai/skills. - Select mantic-mcp, choose Add, then sign in with your Mantic account when prompted.
To add only the MCP server, open Settings → Connectors → Add custom connector, name it Mantic and paste https://api.mantic.com/mcp.
Claude Code
Install the Mantic plugin:
claude plugin marketplace add mantic-ai/skills
claude plugin install mantic-mcp@manticOr add only the MCP server:
claude mcp add --transport http mantic https://api.mantic.com/mcpEither way, run /mcp inside Claude Code to sign in.
ChatGPT
Install Mantic from the ChatGPT marketplace, then sign in with your Mantic account when prompted.
To add the skills as well, install the Mantic plugin in the ChatGPT desktop app:
- Open Plugins and choose Add marketplace → Add from a repository.
- Enter
mantic-ai/skills. - Install mantic-mcp from the Mantic marketplace and sign in with your Mantic account.
Workspace admins can add the marketplace for everyone from Admin → Plugins → Add → Import marketplace, using https://github.com/mantic-ai/skills.
Codex
Install the Mantic plugin, which adds the MCP server and its skills:
codex plugin marketplace add mantic-ai/skills
codex plugin add mantic-mcp@manticIf Codex doesn't prompt you to sign in, run codex mcp login mantic.
Or add only the MCP server:
codex mcp add mantic --url https://api.mantic.com/mcp
codex mcp login manticYou can also configure it directly in ~/.codex/config.toml:
[mcp_servers.mantic]
url = "https://api.mantic.com/mcp"Other MCP clients
Any client that supports Streamable HTTP with OAuth works. A typical JSON configuration, for example in Cursor:
{
"mcpServers": {
"mantic": {
"url": "https://api.mantic.com/mcp"
}
}
}The server advertises its authorization server through standard OAuth discovery, so compliant clients handle sign-in themselves. If a session starts returning authentication errors, sign in again from your client's MCP or connector settings.
Batched forecasts
create_question_set makes a batch of related questions, in one of two kinds:
- A standard batch gives each question one forecast at creation, with no further runs scheduled. You can still refresh each question later.
- A Scout batch tracks each question over time with Scout, a faster, lower-cost version of Mantic's forecaster. Each question gets
runs_per_questionforecasts: the first when the batch is created, and the rest spread evenly untilwindow_end. A Scout batch commits to between 100 and 10,000 runs in total.
Creating a batch takes two steps. create_question_set_draft saves a draft from a title template, its variables (or rows, one per question, where each question has its own values) and the question content every question shares, and returns how many questions it makes with a sample of them; nothing is created yet. After the user reviews the template and the commitment, create_question_set with the draft's ID creates one question for each row, or each combination of the variables' options. A draft creates one batch.
For a Scout batch, give the draft mode: "scout" and runs_per_question, and for repeat runs a future window_end. Where the questions differ in type or answer space, save a question draft for each and pass question_draft_ids to create_question_set instead; each draft's title must fill the template's placeholders.
get_question_set reports progress, get_question_set_questions lists completed forecasts, and get_question_set_runs lists every Scout run and its result. These read tools use the same access rules and forecast-impression billing as the API.