> ## Documentation index > Fetch the complete documentation index at: https://docs.predictefy.com/llms.txt > Use it to discover every available page before exploring further. # AI agents > Connect Predictefy to a chat client or build an agent on the API. **Using an LLM (Claude, Cursor, ChatGPT)?** Predictefy ships an MCP server exposing **43 tools** — **33 read, intelligence, and platform tools** for normalized markets, order books, history, cross-venue intelligence, and trader analytics across **16 served prediction-market venues**, plus **10 guardrailed execution tools** for a non-custodial trade surface. One-line install: `npx -y @predictefy/mcp` (get a free API key at [portal.predictefy.com/keys](https://portal.predictefy.com/keys)). No install? Paste `https://docs.predictefy.com/llms.txt` into your LLM — it is a self-contained integration brief. Building an autonomous agent? Start with the [TypeScript SDK](/guides/sdk/) or [Python SDK](/guides/python-sdk/) and the [published agent skill](/.well-known/agent-skills/predictefy/skill.md). **Pick your route below.** ## Which piece is which | Layer | Asset | Reach for it when | | ------------- | ------------------------------------------------------------------------ | ---------------------------------------------------------------------------------- | | **Context** | [`llms.txt`](/llms.txt) | You want a curated, self-contained integration brief to paste into an LLM. | | **Context** | [`llms-full.txt`](/llms-full.txt) | All 56 authored pages plus the generated API operation catalogue. | | **Context** | Per-page `.md` twins | Clean Markdown for any of the 56 authored pages; API pages have no twin. | | **Tools** | [`@predictefy/mcp`](/guides/mcp/) | A chat client or agent needs callable Predictefy tools at runtime. | | **Tools** | [REST API](/api/) | You need the normalized HTTP contract directly. | | **Tools** | [`@predictefy/sdk`](/guides/sdk/) | You are building a typed TypeScript agent. | | **Tools** | [`predictefy` on PyPI](/guides/python-sdk/) | You are building a Python research, automation, or backtesting agent. | | **Tools** | [`@predictefy/cli`](/guides/cli/) | An agent or operator needs the same API from a shell. | | **Discovery** | [`/.well-known/agent-card.json`](/.well-known/agent-card.json) | An A2A platform needs Predictefy's identity and capability card. | | **Discovery** | [`/.well-known/agent-skills/`](/.well-known/agent-skills/index.json) | A platform needs the published skill URL and integrity digest. | | **Discovery** | [`/.well-known/mcp/server-card.json`](/.well-known/mcp/server-card.json) | An MCP directory or platform builder needs the local stdio install contract. | | **Discovery** | The `llms.txt` convention | A crawler checks the docs origin for an LLM-readable starting point. | | **Discovery** | [`sitemap-index.xml`](/sitemap-index.xml) | A crawler needs the complete index of docs URLs. | ## Route A — use Predictefy from your chat client Start read-first. Use the MCP server for research, market screening, cross-venue comparison, and Trader Intelligence. Treat price gaps as **indicative price discrepancies** unless every live executability gate passes. Create a free key at [portal.predictefy.com/keys](https://portal.predictefy.com/keys), then choose your client. ### Claude Code ```sh claude mcp add predictefy -e PREDICTEFY_API_KEY=pk_live_your_key_here -- npx -y @predictefy/mcp ``` ### Claude Desktop Add this server to `claude_desktop_config.json`: ```json { "mcpServers": { "predictefy": { "command": "npx", "args": ["-y", "@predictefy/mcp"], "env": { "PREDICTEFY_API_KEY": "pk_live_your_key_here" } } } } ``` ### Cursor Create `.cursor/mcp.json` in your project: ```json { "mcpServers": { "predictefy": { "command": "npx", "args": ["-y", "@predictefy/mcp"], "env": { "PREDICTEFY_API_KEY": "pk_live_your_key_here" } } } } ``` ### VS Code Create `.vscode/mcp.json` in your workspace: ```json { "servers": { "predictefy": { "type": "stdio", "command": "npx", "args": ["-y", "@predictefy/mcp"], "env": { "PREDICTEFY_API_KEY": "pk_live_your_key_here" } } } } ``` ### Advanced: trading from chat The **10 execution tools are registered by default**. Set `MCP_ENABLE_TRADE=false` to disable them and run a server that cannot trade or return a collateral-moving transaction to sign. No tool both builds and signs an order: build operations return unsigned artifacts, and signing stays client-side. Server-side spend caps still apply — by default, **100 USD per order** and **1,000 USD per API key over a rolling 24 hours**. Read [Trading & execution](/guides/trading/) before acting; it records the current per-venue reality and guardrails. ## Route B — build an agent on the API Choose context at the granularity your agent needs: - Paste [`llms.txt`](/llms.txt) when the model needs a compact integration brief in-context. - Load [`llms-full.txt`](/llms-full.txt) when you need all 56 authored pages plus the complete generated API operation catalogue. - For one authored page, use `/index.md`, `/quickstart.md`, or replace the trailing slash in a `/guides/**/` or `/reference/**/` URL with `.md`. - Do not append `.md` to `/api/` or `/api/operations/**/`: the generated API reference has no per-page markdown twins. Use [`llms-full.txt`](/llms-full.txt) for its operation catalogue. ### Install the published agent skill The canonical skill is served at `https://docs.predictefy.com/.well-known/agent-skills/predictefy/skill.md`. For Claude Code, install it at `~/.claude/skills/predictefy/SKILL.md`: ```sh mkdir -p ~/.claude/skills/predictefy curl -fsSL https://docs.predictefy.com/.well-known/agent-skills/predictefy/skill.md \ -o ~/.claude/skills/predictefy/SKILL.md ``` For runtimes using the [Agent Skills](https://agentskills.io) ecosystem, install the same file in the shared user directory: ```sh mkdir -p ~/.agents/skills/predictefy curl -fsSL https://docs.predictefy.com/.well-known/agent-skills/predictefy/skill.md \ -o ~/.agents/skills/predictefy/SKILL.md ``` Other runtimes can fetch the raw `.well-known` URL directly. The [skill discovery index](/.well-known/agent-skills/index.json) publishes its URL and digest. ### Discovery cards for platform builders - [`agent-card.json`](/.well-known/agent-card.json) advertises the A2A identity and capabilities. - [`agent-skills/index.json`](/.well-known/agent-skills/index.json) lists the published skill and its integrity digest. - [`mcp/server-card.json`](/.well-known/mcp/server-card.json) describes the local stdio MCP package and install contract; it does not advertise a hosted MCP endpoint. ### Build with the API, SDKs, or CLI - Use the [REST API](/api/) when your agent owns its HTTP client. - Use [`@predictefy/sdk`](/guides/sdk/) for TypeScript or [`predictefy` on PyPI](/guides/python-sdk/) for Python. - Use [`@predictefy/cli`](/guides/cli/), **the terminal and AI-agent client**, when shell commands are the right integration boundary. :::note[Beta releases] These packages are published on beta lines. Pin exact versions rather than tracking `latest`. ::: ```sh npm install @predictefy/sdk pip install predictefy npm install --global @predictefy/cli ```