NEW Promptev MCP Server — One URL. Every tool. Your team’s knowledge. Connect Claude Code, Cursor, or any MCP client. →

Documentation

Everything you need to build, deploy, and manage context-aware AI agents with Promptev.

Quick Start

Getting Started

Create your first project, connect data, and deploy an agent in under 10 minutes.

Read guide

SDK & API

Python & JavaScript SDKs with endpoint reference, schemas, and SSE streaming.

View reference

Tools

233+ built-in system tools, custom HTTP, MCP, and database tools for real-world actions.

Explore tools

Getting Started

Get up and running with Promptev in three steps. No credit card required for the free tier.

1

Create a Project

Sign up at app.promptev.ai and create your first project. Projects are containers for connectors, context packs, prompts, tools, and deployments.

2

Connect Your Data & Models

Connect integrations (Google Drive, Jira, GitHub, etc.) for data and tools. Then add model keys (OpenAI, Claude, Gemini) under BYOK.

Google DriveJiraGitHubSlackNotion+12 more
3

Create a Prompt & Deploy

Build a prompt in Prompt Studio, attach context packs and tools, test in the playground, then deploy as a web agent, API, SDK, or Slack bot.

Context Packs

Context Packs are Promptev's managed knowledge system. They turn your raw data into searchable knowledge that agents use to answer questions accurately.

How It Works

1

Connect data sources — Google Drive folders, Notion spaces, Confluence pages, GitHub repos, uploaded files.

2

Automatic processing — Promptev extracts text, chunks content, generates embeddings, and builds entity graphs.

3

Live re-indexing — Webhook-driven updates keep your context pack in sync when files change.

4

Dual retrieval — Semantic search + graph-based cross-document linking for maximum accuracy.

# Pipeline states
idle → queued → processing → completed

Track status via the dashboard or API. Failed packs show detailed error logs.

Prompt Studio

Version control for AI behavior. Create prompts, attach context packs and tools, test side-by-side, rollback to any version, and deploy with confidence.

Version Control

Every prompt change creates a new version. Compare diffs, rollback, see who changed what.

Side-by-Side Testing

Test two versions against the same input. Compare responses, latency, and token usage.

A/B Testing

Run experiments across models (GPT-4 vs Claude vs Gemini) with traffic splitting.

Audit Trail

Full history of every change, deployment, and test run. Traceable and compliant.

Tools

Tools give agents the ability to act in the real world. Three tool types, one unified governance layer.

System Tools

233+ built-in across 17+ integrations

Pre-built tools that auto-load when you connect an integration. Each has a defined config schema for parameter validation.

HTTP Tools

Custom REST API endpoints

Connect any REST API as a tool. Configure URL, method, headers, auth, and schemas via the dashboard.

GET POST PUT PATCH DELETE

MCP Tools

Model Context Protocol

Connect any MCP-compatible server. MCP is an open standard — think USB-C for AI tool connectivity.

How it works: Add your MCP server URL → Promptev auto-discovers tools → Agents call them with full governance.

Tool Governance

All three types share unified governance: per-tool approval requirements, inline & email notifications, full audit logging, and RBAC.

Deployment

Deploy prompts and agents to any channel. All methods share the same execution engine.

Web AI Agent

Branded shareable link. Optional OTP/email gating for private access.

Embedded Widget

Embed via iframe. Responsive, customizable, works on any domain.

API Endpoint

RESTful API with streaming. Manage conversations programmatically.

Python / JS SDK

Type-safe clients with streaming and async support.

WhatsApp / Slack Bot

Deploy as a bot in WhatsApp or your Slack workspace.

Autonomous Agents

Event-driven, scheduled, or goal-based with persistent memory.

SDK & API

Everything you need to integrate Promptev. Get your project key, install the SDK, and start calling endpoints.

1

Get your Project Key

Go to Project Settings in the dashboard and copy your project API key (pk_...). This key authenticates all calls — no Authorization header needed.

2

Install the SDK

Python
pip install promptev
JavaScript / TypeScript
npm install @promptev/client
View on PyPI → View on npm →
3

API Endpoints

Base URL: https://api.promptev.ai

POST /api/sdk/v1/prompt/client/{project_key}/{prompt_name}

Execute a prompt. Add ?stream=true for SSE streaming.

Request Body

{
  "variables": {
    "key": "value"
  }
}

Response

{
  "output": "The AI response text..."
}
Python
client.run_prompt(prompt_key, query, variables)
client.stream_prompt(prompt_key, query, variables)
JavaScript
client.runPrompt(promptKey, query, variables)
client.streamPrompt(promptKey, query, variables)
POST /api/sdk/v1/agent/{project_key}/{chatbot_id}/start

Start or resume an agent session. Returns a session token for streaming.

Request Body

{
  "visitor": "[email protected]",
  "platform": "sdk"
}

Response

{
  "session_token": "tok_...",
  "chatbot_id": "...",
  "name": "My Agent",
  "memory_enabled": true,
  "messages": []
}
Python
client.start_agent(chatbot_id, visitor, platform)
JavaScript
client.startAgent(chatbotId, { visitor, platform })
POST /api/sdk/v1/agent/{project_key}/{chatbot_id}/stream

Stream an agent response via SSE. Requires session_token from /start.

Request Body

{
  "query": "Create a Jira ticket for this bug"
}
// Pass session_token as query param

SSE Events

{ "type": "thoughts",    "output": "..." }
{ "type": "processing", "output": "..." }
{ "type": "done",       "output": "..." }
{ "type": "error",      "output": "..." }
Python
client.stream_agent(chatbot_id, session_token, query)
JavaScript
client.streamAgent(chatbotId, { sessionToken, query })

Async support (Python): Every method has an async variant — prefix with a (e.g., arun_prompt, astart_agent, astream_agent). JavaScript methods are async by default.

Webhooks

Real-time updates from connected integrations. When files change, issues update, or pages are edited, webhooks trigger automatic context pack re-indexing.

Supported Events

File created / updated / deleted
Jira issue created / updated
GitHub push / PR / issue events
Confluence page created / updated
Slack messages & reactions
Microsoft subscription events

BYOK (Bring Your Own Keys)

Promptev never owns your model keys or data. Connect your own LLM providers and control where inference runs.

Supported Providers

O
OpenAI
C
Claude (Anthropic)
G
Gemini (Google)
M
Mistral
C
Cohere
S
Self-hosted

Security

API keys are encrypted at rest using AES-256, never logged, and never exposed in responses. Rotate or delete keys at any time.

Ready to Build?

Start with the free tier. Deploy your first agent in under 10 minutes.

Start Free Talk to Us

Promptev MCP Server

Connect any MCP-compatible AI tool to Promptev. One URL gives you access to all your project’s tools, team knowledge, and AI agents — no install, no local processes.

Quick Setup

Add this to your AI coding tool’s MCP configuration:

{
  "mcpServers": {
    "promptev": {
      "type": "url",
      "url": "https://api.promptev.ai/mcp/YOUR_PROJECT_API_KEY"
    }
  }
}

Generate your API key in Project Settings → API Keys on app.promptev.ai

What You Get

5 built-in tools that give your AI access to everything — tools, agents, and knowledge — on demand:

  • search_knowledge_base — Search your knowledge bases — documents, files, and data
  • list_tools — Discover all configured tools (233+ system tools + custom HTTP, MCP, and database tools)
  • call_tool — Execute any discovered tool by name with parameters
  • list_agents — Discover all live Promptev agents in the project
  • call_agent — Send messages to agents and get responses

Dynamic discovery: Only 5 tools load into context. Your AI calls list_tools to discover 233+ system tools plus your custom HTTP and MCP tools on demand — keeping context small and responses fast.

Compatible With

  • 💻 Claude Code — claude mcp add-json promptev '{"type":"http","url":"..."}'
  • 💻 Claude Desktop — via mcp-remote bridge (see snippet below)
  • 💻 Cursor — Settings → MCP → Add Server
  • 💻 OpenAI Codex — MCP config
  • 💻 Windsurf — MCP settings
  • 💻 VS Code + Copilot — MCP extension
  • 💻 Any tool that supports the MCP protocol

How It Works

1️⃣
Connect Integrations

OAuth flows in Promptev UI — Google, Jira, Slack, etc.

2️⃣
Index Knowledge

Upload docs or connect cloud storage as context packs

3️⃣
Copy MCP URL

One URL with your project API key

4️⃣
Use Everywhere

List packs, list agents, call agents, run any tool — from Claude, Cursor, Codex, anywhere

Built-in MCP Tools

Five tools are registered on connect. Project tools are discovered on demand via list_tools — no context bloat:

Setup by Client

The MCP URL is the same everywhere — only the config format changes. Pick your tool:

Claude Desktop

Use the built-in Custom Connectors flow — no bridge needed:

  1. Settings → Connectors → Add custom connector
  2. Name: Promptev
  3. URL: https://api.promptev.ai/mcp/API_KEY
  4. Save → toggle on in any chat
Older versions? Use the JSON config ↓

For Claude Desktop versions without Connectors, add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "promptev": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://api.promptev.ai/mcp/API_KEY"
      ]
    }
  }
}

Quit fully (⌘Q) and reopen.

Claude.ai (web)

Native HTTP MCP — no bridge needed. From a chat:

  1. Settings → Connectors → Add custom connector
  2. Name: Promptev
  3. URL: https://api.promptev.ai/mcp/API_KEY
  4. Save → Enable in chat tools panel

Available on Pro, Max, Team, and Enterprise plans.

Claude Code

One command, scoped to your project or user:

claude mcp add-json promptev '{
  "type": "http",
  "url": "https://api.promptev.ai/mcp/API_KEY"
}'

Use --scope user to make it available across all projects.

ChatGPT

Via custom connectors (Pro / Business / Enterprise):

  1. Settings → Connectors → Create
  2. MCP Server URL: https://api.promptev.ai/mcp/API_KEY
  3. Auth: None (key is in URL)
  4. Enable in Developer Mode for full tool access

Or attach to a custom GPT via the GPT builder → Actions → MCP.

OpenAI Codex

Add to ~/.codex/config.toml:

[mcp_servers.promptev]
command = "npx"
args = [
  "-y", "mcp-remote",
  "https://api.promptev.ai/mcp/API_KEY"
]

Restart your Codex session after editing.

Cursor

Edit ~/.cursor/mcp.json (or via Settings → MCP → Add Server):

{
  "mcpServers": {
    "promptev": {
      "url": "https://api.promptev.ai/mcp/API_KEY"
    }
  }
}

Reload Cursor — Promptev tools appear in the agent tool list.

Get Your MCP URL →

Frequently Asked Questions

Create a project, connect your data sources via OAuth, build a Context Pack from your documents, write prompt instructions in plain English, and deploy as a chatbot, API, or Slack bot. No code required — first deploy in under 10 minutes.

PDF, DOCX, PPTX, XLSX, CSV, images (with OCR), code files in 85+ languages, Markdown, HTML, and plain text. Documents are automatically chunked, embedded, and indexed for semantic search.

HTTP tools let you connect any REST API by defining the endpoint, method, headers, and body. MCP tools connect to Model Context Protocol servers for real-time tool discovery. Both support OAuth and approval controls.

Promptev as MCP server: Only 5 built-in tools load into context. Your AI calls list_tools to discover 233+ system tools plus your custom HTTP and MCP tools on demand, then call_tool to execute them. This keeps context small no matter how many tools you configure.

All API keys and OAuth tokens are encrypted at rest using AES-256. Keys are never logged, never exposed in responses, and never used for model training. You can rotate or delete keys at any time from the dashboard.

Yes. The MCP Server is included in all plans, including Free (500 credits/month). Each tool call costs 1 credit. Context pack searches cost 1 credit. No separate pricing for MCP access.

Yes. It uses the standard MCP protocol (HTTP Streamable transport). Works with Claude Code, Cursor, OpenAI Codex, Windsurf, VS Code with GitHub Copilot, and any MCP-compatible tool. Same URL, zero per-tool configuration.

Yes. The MCP URL is scoped to a project. Everyone shares the same tools, context packs, and integrations. New team member? They paste the same URL and get everything from day one.

Yes. Use the built-in list_agents tool to discover agents in your project, then call_agent with an agent_id and message. The agent’s full prompt, attached tools, and context packs execute server-side — the MCP client just sends the message and receives the response.

call_agent returns a session_id with every response. Pass it back on the next call to keep the conversation going — the agent will see the prior messages, retain memory, and respond in context. Omit session_id (or pass an unknown one) to start a fresh conversation.