Agentdevx Sdk

What mirajmahmudul/agentdevx-sdk MCP server does

The mirajmahmudul/agentdevx-sdk MCP server acts as a gateway between an AI agent and registered external APIs. Rather than supplying a separate integration for each service, it lets an API be registered from an OpenAPI specification and then called through the gateway proxy. The gateway associates activity with an agent identity and records tool usage in an audit trail.

Agent identity uses Ed25519 keys instead of OAuth tokens. Credentials are stored in an AES-256-GCM encrypted vault and scoped to agents. The gateway also provides encrypted memory that persists between sessions, an OPA-based policy layer, capability tags, and a dry-run mode for policy decisions. The README describes more than 42 pre-registered tools in addition to custom OpenAPI registrations.

How it works

An agent first bootstraps through POST /agents/v1/bootstrap, sending a name, email, and public key. The response includes a JWT, 75,000 credits, and access to registered tools. After authentication, the agent can inspect its profile and credit balance, read or write memory, save API credentials, and invoke tools through POST /proxy/call.

MCP clients communicate with the gateway at https://agentdevx.onrender.com/mcp using the MCP JSON-RPC endpoint. Credential injection allows stored API keys to be supplied when the gateway proxies a registered call. The policy engine evaluates access, while the audit log records both permitted and rejected actions.

The mirajmahmudul/agentdevx-sdk MCP server is suited to agents that need a stable identity and shared gateway controls across multiple API calls. It is also relevant when an agent must retain encrypted state across sessions or discover integrations through a single MCP connection.

Setup and configuration

The documented installer is available through npm:

npx @agentdevx/install

The README then shows an MCP configuration that points a client at the hosted endpoint:

{
  "mcpServers": {
    "agentdevx": {
      "url": "https://agentdevx.onrender.com/mcp"
    }
  }
}

The repository does not specify required environment variables for this hosted setup. The bootstrap example uses OpenSSL to generate a public key before making the registration request. The documentation names Claude Desktop, Cursor, Windsurf, and MCP-compatible clients as supported connection options.

Tools and capabilities

Available gateway operations described by the API reference include:

• Create an agent and obtain authentication credentials through /agents/v1/bootstrap.
• Retrieve the current agent profile and credit balance from /agents/me.
• Read and write encrypted per-agent memory through /agents/me/memory.
• Execute registered tools with /proxy/call.
• Store encrypted API keys with /credentials.
• Create a billing checkout through /billing/checkout.
• Use the /mcp JSON-RPC endpoint from an MCP client.
• Register APIs from OpenAPI specifications and call them through the proxy.

The security model combines Ed25519 agent identity, encrypted credential storage, OPA policy evaluation, capability tags, dry-run checks, and an immutable audit record. The listed pricing includes a free plan with 100 calls per day and 75,000 credits, although limits and plan details may change outside the repository documentation.

Limitations and notes

The provided material documents a hosted gateway URL and an npm installer, but it does not describe local deployment steps, a complete list of MCP tool names, or the exact request schema for every endpoint. It also does not list environment variables required by the server. API access depends on the relevant service being registered and on credentials and policy rules allowing the call.

The mirajmahmudul/agentdevx-sdk MCP server is therefore best evaluated as a gateway and control layer, not as a fixed catalog of individually documented service integrations. The README states that self-bootstrapping is supported, but it does not provide implementation details for key management, policy authoring, or OpenAPI registration beyond the stated capabilities.