RocketPool Copilot: MCP Server and Guarded AI Agent for Node Operators
What is the work being proposed?
I’m proposing to build an MCP (Model Context Protocol) server for Rocket Pool node operators. MCP is the open standard that lets AI assistants like Claude connect to real data and take real actions through defined tools, instead of just talking about them.
Right now, checking on a Rocket Pool node means reading the Smart Node CLI output, cross-referencing a Grafana dashboard, and knowing what RPL collateral ratios and penalty thresholds actually mean. That’s a real barrier. The Node Operator Experience section of this forum has said as much directly: people find node operation intimidating and full of unknowns, and that stops them from trying.
RocketPool Copilot turns that into a conversation. A node operator, or someone thinking about becoming one, could ask things like “why is my minipool at risk,” “what’s my current RPL collateral ratio,” or “walk me through creating a minipool,” and get a straight answer pulled from live node and chain data. For a defined set of safe actions such as claiming rewards or topping up RPL, the agent can also execute them directly, but only through a policy engine that blocks anything outside a strict allow-list: no unknown contracts, no unsafe calls, spend caps enforced before anything is broadcast. Every decision the agent makes, allow or deny, gets written to an audit log so nothing happens silently.
Is there any related work this builds off of?
Yes. I built and shipped the same kind of system for Rootstock: an MCP server with five tools (protocol data, swap simulation, intent execution, position health, wallet balances) sitting behind a policy engine that blocks zero-address transfers, selfdestruct calls, oversized single transactions, and non-whitelisted contracts, with every allow or deny decision logged. It’s live on Rootstock mainnet today, repo is Rootstock-Agentic-DeFi-Framework under my GitHub: soloking1412 (Maheswaran Velmurugan) · GitHub
RocketPool Copilot is that same architecture, rebuilt against Rocket Pool’s own data sources (Smart Node API, Beacon chain, rETH and RPL contracts) instead of Rootstock’s. I’m not starting from a blank page on the hardest part, which is the safety layer.
Will the results of this project be entirely open source (MIT, GPL, Apache, CC BY license or similar)? If not, which parts will not be, why, and under what license will they be published?
Yes, MIT license, published on GitHub from day one of development, not just at the end.
Benefit
| Group | Benefits |
|---|---|
| Potential rETH holders | N/A, this tool is aimed at the node side of the protocol, not rETH holders directly |
| rETH holders | Indirectly benefits from a healthier, better-monitored node operator base, since minipool health affects rETH’s backing |
| Potential NOs | Lowers the barrier to becoming a node operator by replacing dashboard-reading and CLI familiarity with plain-language diagnostics and guided onboarding |
| NOs | Gives existing operators a faster way to check collateral ratios, penalty risk, and rewards status, and to safely automate routine actions like claiming rewards or topping up RPL |
| Community | Adds a genuinely new piece of open source tooling to the ecosystem that didn’t exist before, and one that’s reusable by anyone building on Rocket Pool data |
| RPL holders | Better collateral monitoring means fewer operators getting caught under-collateralized, which is good for RPL’s role as the collateral asset |
Which other non-RPL protocols, DAOs, projects, or individuals, would stand to benefit from this grant?
The MCP server pattern itself is reusable. Any team building AI-agent tooling on top of liquid staking or validator infrastructure could fork the policy engine approach. I’ve already proven that by taking it from Rootstock to here.
Work
Who is doing the work?
Me, solo. I’m Maheswaran Velmurugan (Soloking), a full stack and blockchain developer based in Tamil Nadu, India. My GitHub is linked above. Telegram: @maheswar1412.
What is the background of the person(s) doing the work? What experience do they have with such projects in the past?
I’m a Polkadot Blockchain Academy graduate and a Polkadot Technical Fellow. I build end to end, from Solidity and Clarity contracts to Python SDKs to TypeScript CLI tools, and I ship as a solo developer without a team.
I’ve completed grant-funded projects for both the Arbitrum Foundation and the Stacks Foundation:
- Stylus-Toolkit (Arbitrum Foundation grant, completed): a CLI development environment for Arbitrum Stylus contracts. Both milestones delivered and closed out. Live on NPM with over 200 downloads, with a benchmark suite showing 32.6% average gas savings over equivalent Solidity.
- satoshi-yield (Stacks Foundation grant, completed): a non-custodial sBTC yield optimizer on Stacks, comparing live APY across five protocols with one-call rebalancing.
And directly relevant to this proposal, two projects rewarded through Rootstock’s Hacktivator marketplace:
- Rootstock Agentic DeFi Framework: the MCP server and policy engine this proposal is modeled on, live on Rootstock mainnet, rewarded via Hacktivator
- Rootstock Python SDK: published on PyPI, wallet management, ERC-20, RNS resolution, also rewarded via Hacktivator
All of these are visible on my GitHub linked above. Plus security research: confirmed findings across Sherlock, Code4rena, and Immunefi, including high-severity findings on protocols like Threshold Network and SuperDCA.
Both the Arbitrum and Stacks grants ran through to full milestone completion, and both Rootstock submissions were reviewed and rewarded on their own merits. That’s a track record of finishing what I propose, across four separate ecosystems. The security research matters specifically here too, since this proposal’s core piece is a policy engine that has to correctly refuse unsafe actions, and that’s the same class of bug I’ve spent real time hunting in other people’s contracts.
What is the breakdown of the proposed work, in terms of milestones and/or deadlines?
Milestone 1 (3 weeks): Read-only diagnostics core
- MCP server scaffold connecting to Smart Node API and Beacon chain data
- Tools: node status, minipool health, RPL collateral ratio, penalty risk, pending rewards
- Basic conversational onboarding flow for prospective node operators
- Public GitHub repo, MIT licensed, with setup docs
Milestone 2 (3 weeks): Guarded actions and policy engine
- Policy engine ported and adapted from the Rootstock implementation: whitelisted contracts only, spend caps, unsafe call blocking
- Guarded action tools: claim rewards, RPL top-up, node fee voting
- Full audit log of every allow and deny decision
- End to end testing against Hoodi testnet
- Final documentation and a short demo video
How is the work being tested? Is testing included in the schedule?
Yes, testing is part of both milestones, not an afterthought. Read-only tools get tested against real mainnet data during milestone 1. Guarded actions get tested against Hoodi testnet in milestone 2 before anything touches mainnet, with specific test cases for the policy engine’s refusal paths (the unsafe actions it’s supposed to block), not just the happy path.
How will the work be maintained after delivery?
I maintain all of my grant-funded projects after delivery. Stylus-Toolkit is still live and maintained a year after its Arbitrum grant closed. I’ll do the same here, keeping the tool working as Smart Node’s API evolves.
Costs
What is the acceptance criteria?
Milestone 1: a working MCP server that answers real diagnostic queries against live Rocket Pool data, open sourced with setup documentation.
Milestone 2: guarded write actions working end to end on Hoodi testnet with a demonstrable, logged policy engine, plus final documentation.
What is the proposed payment schedule for the grant? How much USD $ and over what period of time is the applicant requesting?
$8,000 total, split as $4,000 on delivery of Milestone 1 and $4,000 on delivery of Milestone 2, over roughly 6 weeks.
Who will directly receive the payment? (Required — the GMC can only send grants directly to a recipient address and cannot accept invoices.)
0x2be8023cB03428e988Fa3Ad4079fcf23673c2595
How will the GMC verify that the work delivered matches the proposed cadence?
Each milestone ships as a public commit history on GitHub, so progress is visible throughout, not just at handoff. I’ll also post a short update in the discussion thread at each milestone with what’s working and a way to try it.
What alternatives or options have been considered in order to save costs for the proposed project?
I considered scoping this down to read-only diagnostics only and dropping the guarded actions milestone, which would cut the ask roughly in half. I kept both in because the guarded actions are what actually reduces friction for operators day to day, not just informs them, and the policy engine reuse from Rootstock keeps that milestone’s cost down since I’m not designing that part from scratch.
Have you already been compensated by the RP protocol in any way for this work?
No.
Conflict of Interest
Does the person or persons proposing the grant have any conflicts of interest to disclose? (Please disclose here if you are a member of the GMC or if any member of the GMC would benefit directly financially from the grant).
None. I’m not a member of the GMC and have no prior relationship with Rocket Pool.
Will the recipient of the grant, or any protocol or project in which the recipient has a vested interest (other than Rocket Pool), benefit financially if the grant is successful?
No.