RPCX MCP Server
Connect a coding agent to the RPCX MCP server so it can read the live tables, the rules, and the method language, and validate methods before you run them.
The RPCX MCP server gives your coding agent what it needs to write a correct method: the nine live tables and their columns, the rules for reading them in bulk, the method language with example methods, and a validator. It uses the Model Context Protocol, so it works with Claude Code and other MCP clients.
The MCP server never runs a method and never bills you. Methods run only from the dashboard or as rpcx_<name> calls with your API key.
The server is at https://rpcx.mainnet.solana.nolimitnodes.com/mcp.
Get a key
Open the RPCX page in the dashboard. In Use with an AI agent, select Generate key. Copy the key when it appears. It starts with rpcx_mcp_ and is shown once.
The MCP key is not your API key. It only lets an agent read the tables and validate methods. Each workspace has one active key. Regenerate creates a new key and revokes the old one. Revoke stops the key. An agent with a revoked key gets HTTP 401, within about a minute.
Treat the key like a password. Do not commit it to a repository.
Connect Claude Code
The RPCX page shows this command with your key filled in:
claude mcp add --transport http rpcx https://rpcx.mainnet.solana.nolimitnodes.com/mcp \ --header "Authorization: Bearer rpcx_mcp_YOUR_KEY"Connect other clients
The server uses the Streamable HTTP transport with JSON responses. Send the key as a bearer token:
{ "mcpServers": { "rpcx": { "type": "http", "url": "https://rpcx.mainnet.solana.nolimitnodes.com/mcp", "headers": { "Authorization": "Bearer rpcx_mcp_YOUR_KEY" } } }}The server accepts MCP protocol versions 2025-06-18, 2025-03-26, and 2024-11-05.
Tools
When the server starts a session, it also sends the correctness rules as instructions, so the agent sees them before it writes anything. The server only teaches the method language. Raw SQL is refused.
The server answers JSON, not a stream, and keeps no session. A check with test values is limited to 30 per key per minute; past that, validate_method still runs the static checks and adds a dry_run_rate_limited warning.
Validate before you paste
Ask the agent to validate every method. validate_method runs the same checks as the dashboard. It also checks the method on empty tables, which catches problems such as a column that does not exist. The check reads no data from Solana and is not billed. It does not return a cost estimate.
When a method is valid, the result includes call_as (rpcx_<name>) and a copy_paste block. The block tells you what to do in the dashboard, lists the test values the agent used, and holds the method text. The agent is told to end its answer with that block unchanged.
To use it, open the RPCX page, select New Method, and paste the method text into the Code tab. The name and params fill in. Enter the test values in Params, select Save, then Run.
What to ask for
- Say which addresses the method reads and whether you want to pass many at once. The agent makes a list param such as
wallets: pubkey[]. - Say the time range you care about, or ask for current balances and holdings. A method that reads by address should always take a time range: ask for
fromandtoparams. - Say if an address is very busy, such as a program or a popular mint. The agent can then read a narrower range or a cheaper table, such as
live_signatures. - Say which values you want to pass when you call the method, for example a token mint or a time window. The agent makes those params.
- Check the result of a new method before you rely on it.