MCP Klever Vm

What klever-io/mcp-klever-vm MCP server does

The klever-io/mcp-klever-vm MCP server provides development and blockchain tools for the Klever network. Its knowledge-base tools search, retrieve, rank, and augment Klever VM guidance, including code examples, best practices, security tips, scripts, and documentation. Developers can use search_documentation for formatted answers to implementation questions, or use query_context, get_context, and find_similar when they need structured entries and related material.

The server also inspects and interacts with Klever data. It can return KLV or KDA balances, account state, asset configuration, transaction details, block information, and active validators. query_sc performs read-only smart contract calls and returns base64-encoded results. analyze_contract checks Rust source for several common Klever contract problems, including missing macros, endpoint annotations, storage mapper attributes, payable-handler usage, and event definitions.

Transaction-building tools cover transfers, contract deployment, contract invocation, and KLV freezing. These tools produce unsigned transaction data; signing remains an external client responsibility.

How it works

The server can operate as an MCP stdio process, an HTTP service, or a public hosted service. Local MCP clients can launch the npm package with npx, while the hosted endpoint is available at https://mcp.klever.org/mcp. In local deployments, knowledge is loaded into memory by default. Redis can be selected when context should persist across restarts.

Knowledge entries can be added and later searched through MCP. The server also includes project-oriented tools: init_klever_project scaffolds a Rust contract project, add_helper_scripts adds automation scripts to an existing project, and SDK status or installation tools manage the Klever development prerequisites.

Setup and configuration

The klever-io/mcp-klever-vm MCP server can be started locally with npx -y @klever/mcp-server. Source-based setup uses pnpm install, followed by pnpm run build; MCP mode is started with MODE=mcp pnpm start. Claude Desktop and Cursor configurations use the same stdio command and package arguments.

Configuration includes MODE, PORT, and STORAGE_TYPE. Memory storage supports a MEMORY_MAX_SIZE limit. Redis storage uses REDIS_URL and requires the knowledge-base ingestion step before startup. Public deployments can also configure CORS, request rate limits, and request body size. Contract transaction workflows require the Klever SDK tools, which can be installed into ~/klever-sdk/ using the included SDK installer.

Tools and capabilities

• Search and enrich Klever VM development context.
• Inspect accounts, balances, assets, transactions, blocks, and validators.
• Query read-only contract state and analyze Rust contract source.
• Build unsigned transfers, deployments, invocations, and freeze transactions.
• Create projects, add helper scripts, and check or install SDK components.

Limitations and notes

The klever-io/mcp-klever-vm MCP server does not sign transactions or handle private keys. Contract invocation and deployment therefore require an external signing workflow. Read-only query_sc calls should be used instead of state-changing invocation when no transaction is needed.

The public hosted mode exposes only a read-only subset of the tools. In that mode, context creation and shell-based project tools are disabled. Memory storage is not persistent, whereas Redis storage is intended for data that must survive restarts. Raw asset response fields such as IDs, names, and tickers may be base64-encoded, and contract call arguments and return data use base64 encoding.