
Reactive MCP is a Model Context Protocol server that answers questions about reactive contracts (RCs) using live chain data. Connect it to Claude, Cursor, VS Code, or ChatGPT, then ask in plain language: why a contract didn’t react, whether its subscriptions are active, whether it’s in debt, why a callback never arrived, and what it will cost to run.
Reactive MCP is read-only. It never asks for private keys and never sends transactions. The endpoint is public and needs no authentication. Reactive MCP is an early release (v0.1), so expect tools and report formats to evolve.
https://mcp.rnk.dev/mcpConnecting
Claude Code
claude mcp add --transport http reactive-mcp https://mcp.rnk.dev/mcpNetworks
Every tool takes an optional network parameter: lasna (default) or mainnet. Aliases work too: testnet or 5318007 for Lasna, 1597 for mainnet. In practice, just name the network in your question, as in “check 0x… on mainnet”.
Tools
| Tool | What It Answers |
|---|---|
get_network_info | Chain ID, system contract, cron emitter, supported origin and destination chains with their callback proxies, node capabilities, fee parameters, and whether the sequencer is ingesting origin-chain events right now. |
list_subscriptions | What an RC is subscribed to and whether each subscription is active, from the live filter index and Subscribe/Unsubscribe history. Pass the deployer address to catch subscriptions made from the deployer EOA by mistake. |
validate_subscription | Checks subscribe() criteria before or after deployment: chain support, mainnet/testnet mixing, topic0 vs. event signature (uint instead of uint256 is the classic), topics beyond the event’s indexed parameters, zero topics used as wildcards instead of REACTIVE_IGNORE, origin contract code, matching events per day, and whether the filter is registered. |
get_rc_health | Whether the RC can react and pay right now: debt, reserves, balance, active subscriptions, recent Blacklist/Whitelist/PaymentFailure events, and react() reverts with decoded errors. |
trace_origin_event | Why an RC did or didn’t react to an origin-chain transaction. Walks origin receipt → filter match → sequencer → reactive transaction → callback request → delivery, and stops at the stage that broke. |
trace_reactive_tx | Explains a Reactive transaction (for example, one found on Reactscan): which origin log triggered it, whether react() reverted, which callbacks it requested, and what’s known about their delivery. |
decode_callback | Checks a callback request from a reactive transaction hash or a raw payload: the address slot in the first argument, gas limit against the callback proxy’s overhead, recipient code, _CALLBACK_SENDER, and destination debt. |
estimate_cost | REACT per reaction and per day (events and cron ticks), plus destination-chain callback cost. Returns USD figures only if you pass prices. |
explain_error | Decodes revert data or an error name/selector (InDebt, NotAuthorized, CallbackNotAuthorized, InsufficientGas, Panic, and others). Works offline. |
Resources and Prompts
The server also exposes reference resources the tools draw on:
| Resource | Contents |
|---|---|
rnk://pitfalls | Common reactive contract mistakes |
rnk://errors | Error names, selectors, and meanings |
rnk://events | Reactive Network system events |
rnk://networks | Supported networks and their parameters |
Two prompts package common workflows:
diagnose_rcruns a step-by-step diagnosis: health → subscriptions → trace.review_rc_sourcereviews RC source code against a checklist of common mistakes.
Reading the Reports
Every tool answer is a report with the same structure:
- Verdict: a short code such as
OK_SUBSCRIBED,LOW_RESERVES,TOPIC0_MISMATCH, orDEST_IN_DEBT. - Confidence:
proven(backed directly by chain data),likely(strongly indicated but not conclusive), orcannot_prove(the data available can’t settle it). - Findings: the evidence, with transaction hashes, block numbers, and explorer links.
- Fix: what to change.
- Limits: what the tool couldn’t check.
Example Questions
You don’t need to name tools. The agent picks them. Questions like these work well:
- “Why isn’t my contract at
0x…reacting to Transfer events on Sepolia?” - “Is my subscription correct for
Swap(address,address,int256,int256,uint160,uint128,int24)?” - “Why did callbacks to Sepolia stop arriving?”
- “How much REACT will this contract burn per day?”
- “What does revert data
0x…mean?”
A question like “why no callbacks on Sepolia?” typically chains list_subscriptions, get_rc_health, and trace_reactive_tx. A common answer: the RC reacts fine, but the callback recipient owes debt at the Sepolia callback proxy.
Rate Limits
The endpoint is shared and rate-limited. When it’s busy, you’ll see one of these:
- HTTP
429or503: too many requests. Retry shortly. SERVER_BUSY: the tool call wasn’t run. The result includes a retry hint.RPC_LIMIT_REACHED: the server couldn’t get an RPC slot within the call’s deadline. Retry shortly.