clevername-mcp

The official MCP server for Clevername — AI agent governance, security, and routing.

No account needed to get started. Run one command, call one tool, and you're in.

Quickstart (2 steps)

Step 1 — Add the server to your client

Claude Code

claude mcp add clevername-mcp --transport http https://clevername.net/api/hub/mcp/gateway

Cursor — one-click install:

Or add manually to your Cursor MCP config:

Claude Desktop / Windsurf — add to your MCP config:

{
  "mcpServers": {
    "clevername": {
      "command": "npx",
      "args": ["-y", "@clevername/clevername-mcp"]
    }
  }
}

Step 2 — Log in from the chat

Ask your AI assistant:

"Call the clevername_login tool"

You'll see a code like WXYZ-5678 appear. Open https://clevername.net/device, enter the code, then sign in with Google or your email and password — done. Your token is saved automatically and every tool unlocks.

That's it. No dashboard visit required before logging in. No CLEVERNAME_API_KEY to copy and paste.

How it works

Your MCP client (Claude Code, Claude Desktop, Cursor…)
  └── @clevername/clevername-mcp  (local stdio process)
        ├── clevername_login       ← device flow, no pre-auth needed
        └── all other tools        ← unlocked after login
              └── https://clevername.net/api/hub  (hub-core)
                    └── Your connected MCP servers, AI models, memory, audit

The clevername_login tool uses the OAuth 2.0 Device Authorization Grant (RFC 8628). It works even before you have an account — signing in at /device creates your account automatically via Google OAuth.

Available tools

Authentication

| Tool | What it does | |------|-------------| | clevername_login | Authenticate from the CLI — prints a device code, polls until you approve at clevername.net/device. Works with no pre-existing token. |

Identity & Health

| Tool | What it does | |------|-------------| | get_identity | Who you're logged in as | | list_models | Available AI models | | health_check | Hub status | | refresh_tools | Force-refresh the tool list |

AI Chat

| Tool | What it does | |------|-------------| | chat_complete | Chat with any model via your BYOK keys | | route_plan | Plan a multi-step agentic task |

Multi-provider LLM (BYOK)

| Tool | What it does | |------|-------------| | ask_claude | Call Anthropic with your key | | ask_gpt | Call OpenAI with your key | | ask_gemini | Call Google with your key | | ask_llm | Call any provider by name | | ask_best | Route to the best model for the task | | ask_consensus | Ask multiple models and compare answers |

Memory

| Tool | What it does | |------|-------------| | create_memory | Save something to persistent memory | | list_memories | List saved memories | | search_memories | Search by content | | delete_memory | Delete a memory |

MCP Management

| Tool | What it does | |------|-------------| | list_mcp_connections | See your connected MCP servers | | add_mcp_connection | Connect a new MCP server | | call_mcp_tool | Call a tool on any connected server | | list_mcp_tools | List tools from a server | | ping_mcp_connection | Check if a server is reachable | | delete_mcp_connection | Remove a server |

MCP Activation (plugin registry)

| Tool | What it does | |------|-------------| | list_available_mcps | Browse the MCP plugin registry | | activate_mcp | Enable a plugin (GitHub, Slack, Notion…) | | deactivate_mcp | Disable a plugin | | list_active_mcps | See what's enabled |

Guard (AI agent security)

| Tool | What it does | |------|-------------| | guard_onboard_agent | Register an AI agent with guardrails and get a secure token | | guard_list_agents | List registered agents | | guard_get_profile | Get a compiled guardrail profile | | guard_register_agent | Register without full onboarding | | guard_revoke_token | Revoke an agent's token |

Opt-in groups

Enable additional tool groups with set_tool_groups:

| Group | Tools | |-------|-------| | telemetry | Usage summaries, cost reports, audit logs | | projects | Project CRUD | | config | Export config, sync from Claude Desktop | | aithroyz | Cybersecurity environment provisioning | | logging | Query GCP logs, error monitoring |

Advanced setup

Use an existing token (CI/CD, headless environments)

If you're running in a headless environment or already have a token from the dashboard:

{
  "mcpServers": {
    "clevername": {
      "command": "npx",
      "args": ["-y", "@clevername/clevername-mcp"],
      "env": {
        "CLEVERNAME_API_KEY": "cnk_your_token_here"
      }
    }
  }
}

Tokens are available at clevername.net/dashboard/settings/developer.

Connect directly via HTTP (modern clients)

If your client supports HTTP MCP with OAuth:

https://clevername.net/api/hub/mcp/gateway

This skips the local process entirely — the hub handles auth via browser OAuth.

Troubleshooting

"clevername_login" times out — The device code expires after 5 minutes. Call the tool again to get a fresh code.

Tools show "Not authenticated" — Call clevername_login first. If you already logged in, check that ~/.clevername/config.json exists.

npx not found — Install Node.js (v18+) and restart your terminal.

Tools don't appear after login — Run refresh_tools or restart your MCP client.

"Unauthorized" on a specific tool — Your token may have been revoked. Call clevername_login again to get a fresh one.

Development

cd integrations/clevername-mcp
npm install
npm run build
node dist/index.js   # stdio mode

License

MIT