MCP Setup

Promptmark includes a built-in MCP server that lets AI assistants manage your prompt library directly. Connect any MCP-compatible client over Streamable HTTP.

Connection Details

Client Configuration

Most MCP clients handle authentication automatically. You provide the server URL, and the client walks you through a one-time login via your browser. No tokens to copy.

claude mcp add --transport http promptmark https://promptmark.ai/mcp

{
  "mcpServers": {
    "promptmark": {
      "type": "url",
      "url": "https://promptmark.ai/mcp"
    }
  }
}

{
  "mcpServers": {
    "promptmark": {
      "url": "https://promptmark.ai/mcp"
    }
  }
}

{
  "servers": {
    "promptmark": {
      "type": "http",
      "url": "https://promptmark.ai/mcp"
    }
  }
}

{
  "promptmark": {
    "url": "https://promptmark.ai/mcp"
  }
}

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "promptmark": {
      "type": "remote",
      "url": "https://promptmark.ai/mcp",
      "enabled": true
    }
  }
}

For agent frameworks (CrewAI, LangChain, Pydantic AI, OpenClaw), see Agent Frameworks below.

How OAuth Works

When you add the Promptmark URL to your MCP client, authentication happens automatically behind the scenes. Here is what the client does:

1. The client discovers Promptmark’s auth endpoints via:

   • GET /.well-known/oauth-protected-resource (RFC 9728)
   • GET /.well-known/oauth-authorization-server (RFC 8414)

2. The client registers itself dynamically via POST /oauth/register (RFC 7591) and initiates an authorization code flow with PKCE.

3. Your browser opens to the Promptmark login page. You log in with Civic Auth and approve the connection.

4. The client exchanges the authorization code for a token via POST /oauth/token and uses it for all subsequent requests.

Device Flow (Alternative)

Some clients use the OAuth Device Flow instead of the browser redirect flow. The result is the same — you log in and approve — but the mechanism differs:

1. The client requests a device code:

{
  "device_code": "abc123...",
  "user_code": "ABCD-1234",
  "verification_uri": "https://promptmark.ai/auth/device",
  "expires_in": 900,
  "interval": 5
}

2. You visit the verification_uri in your browser, log in, and enter the user_code to approve.

3. The client polls for a token:

{
  "access_token": "eyJhbG...",
  "token_type": "Bearer",
  "expires_in": 86400
}

4. The client uses the returned token as Authorization: Bearer <token> for all subsequent requests.

Bearer Tokens

For agents, scripts, CI/CD pipelines, and clients that don’t support OAuth, you can authenticate with a static bearer token.

1. Log in to Promptmark at promptmark.ai
2. Go to Settings > Developer
3. Copy your Bearer token from the Bearer Token section
4. Pass it in the Authorization header: Authorization: Bearer YOUR_TOKEN

To use a bearer token in a client config file, add the headers field:

{
  "mcpServers": {
    "promptmark": {
      "type": "url",
      "url": "https://promptmark.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Agent Frameworks

Agent frameworks connect to MCP servers programmatically. These use bearer tokens rather than OAuth — generate a token from Settings > Developer in Promptmark before proceeding.

CrewAI

from crewai.mcp import MCPServerHTTP

mcps = [
    MCPServerHTTP(
        url="https://promptmark.ai/mcp",
        headers={"Authorization": "Bearer YOUR_TOKEN"},
        streamable=True,
    ),
]

LangChain / LangGraph

from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "promptmark": {
        "transport": "http",
        "url": "https://promptmark.ai/mcp",
        "headers": {
            "Authorization": "Bearer YOUR_TOKEN"
        },
    }
})
tools = await client.get_tools()

Requires langchain-mcp-adapters v0.2.0+.

OpenClaw

openclaw mcp set promptmark '{"url":"https://promptmark.ai/mcp","transport":"streamable-http","headers":{"Authorization":"Bearer YOUR_TOKEN"}}'

Or in config under mcp.servers. Manage with openclaw mcp list, openclaw mcp show, openclaw mcp unset.

Pydantic AI

import httpx
from pydantic_ai import Agent
from pydantic_ai.mcp import MCPServerStreamableHTTP

server = MCPServerStreamableHTTP(
    "https://promptmark.ai/mcp",
    http_client=httpx.AsyncClient(
        headers={"Authorization": "Bearer YOUR_TOKEN"},
    ),
)

agent = Agent("openai:gpt-4o", toolsets=[server])

async def main():
    result = await agent.run("List my prompts in Promptmark")
    print(result.output)

Requires pydantic-ai v0.2.0+ and httpx.

Other Frameworks

These frameworks also support Streamable HTTP MCP connections with bearer token auth:

• Google ADK — McpToolset with StreamableHTTPConnectionParams (docs)
• Vercel AI SDK — createMCPClient() in @ai-sdk/mcp v6+ (docs)
• Microsoft AutoGen — StreamableHttpMcpToolAdapter in autogen-ext (docs)
• Mastra — MCPClient from @mastra/mcp (docs)
• n8n — MCP Client Tool node with HTTP Streamable transport (docs)

Verifying the Connection

Once connected, ask your AI assistant to list your prompts:

“List my prompts in Promptmark”

The assistant should call the list_prompts tool and return your prompt library. If you see an authentication error, check that your OAuth session is active or verify your token in Settings > Developer.

Available Tools

After connecting, your AI assistant has access to 74 tools across these categories:

See MCP Reference for detailed documentation of every tool, including parameters, types, and examples.