# Connect your AI

Connect Claude, ChatGPT, Grok or Cursor to Prospy over MCP, and let your agent research threads and draft replies you post yourself.

Prospy has a remote MCP server, so the AI assistant you already use can read your scored, ranked threads, research them and draft replies in your voice. You review every reply and post it from your own account: Prospy has no tool that posts, comments, votes or sends messages.

## The workflow
1. **Prospy finds, scores and ranks.** New posts from your keywords, scored for relevance and buying intent, with the Google position of threads that rank. See [Your feed](/docs/feed).
2. **Your agent reads and drafts.** It calls `find_reply_opportunities`, reads the threads and drafts replies that answer the question first and disclose that you work on your product.
3. **You review and post.** Edit the words, check the community’s self-promotion rules and post as yourself.

## Add the server
Server URL: `https://prospy.io/api/mcp` (Streamable HTTP).

- **Claude Code**: `claude mcp add --transport http prospy https://prospy.io/api/mcp`. Run in your terminal, then sign in when asked.
- **Claude**: `https://prospy.io/api/mcp`. Settings → Connectors → Add custom connector, then paste the URL.
- **ChatGPT**: `https://prospy.io/api/mcp`. Settings → Connectors → Create, then paste the URL.
- **Cursor**: `{ "mcpServers": { "prospy": { "url": "https://prospy.io/api/mcp" } } }`. Add to .cursor/mcp.json.
- **Grok and others**: `https://prospy.io/api/mcp`. Any client that supports remote MCP servers: paste the URL and sign in, or send an API key as a bearer token.

Your client opens a Prospy sign-in and asks you to approve access (OAuth 2.1 with PKCE). Prefer a key? Create one in [Integrations → API & MCP](/integrations/api) and send `Authorization: Bearer <key>`. Never paste a key into a shared chat. Signed in, [Integrations → AI agent](/integrations/agent) has the setup for each client and ready-to-paste prompts.

## Prompts
The server ships prompts that run the workflow. Clients that support MCP prompts list them; in Claude Code they are slash commands such as `/mcp__prospy__reply-plan`.

| Prompt | Arguments | What it does |
| --- | --- | --- |
| `reply-plan` | `projectId`, `days`, `limit` (all optional) | Pulls high-intent threads and threads ranking on Google, summarises each question, outlines a helpful, disclosed reply, flags community rules and writes the top 3 in your voice. |
| `daily-brief` | `projectId` (optional) | What is new since yesterday, the five threads to act on, threads ranking on Google and anything negative. |
| `reply-to-thread` | `mentionId` | One thread: what they ask, whether a reply would help, and a reply in your voice. The robot button on a post in your feed copies the same prompt. |

## Try asking
- “Plan my replies for this week.”
- “Which threads rank on Google for my keywords, and what could a reply add?”
- “Which mentions this week show buying intent?”
- “Add ‘alternative to Brand24’ as a keyword on Reddit and X.”

## Tools
Each tool has the same name and arguments as an [API](/docs/api) operation, plus `find_reply_opportunities`, which exists only on MCP. Start with `list_projects` (or `create_project` if you have none), then `find_reply_opportunities` or `list_mentions` with a `projectId`. Path and query values are top-level arguments; fields for writes go inside `input`.

`find_reply_opportunities` takes `projectId`, `days` (default 7), `limit` (default 10) and `minLeadScore` (default 40). It returns `highIntent` and `rankingOnGoogle`: open and follow-up threads with their URL, platform, community, lead score, intent, matched keyword and `google.position` when they rank.

```json
{"name":"list_mentions","arguments":{"projectId":"00000000-0000-4000-8000-000000000001","limit":25,"ranksOnGoogle":true,"sort":"google_rank"}}
```

Results are JSON under `result`. Errors set `isError` with a code and message. Read-scope keys only see read tools; writes need a full-scope credential and the editor role.