01
Open source, not a closed API
Kaskad's MCP server is fully open source under MIT. No API key gating, no rate-limit tier, no proprietary lock-in. Fork it, audit it, run it next to your agent.
Agentic DeFi
Lending for humans — and for autonomous agents.
Kaskad ships with an open-source MCP server. Connect any agent runtime — Claude, GPT, OPC — and let it supply, borrow, monitor, and rebalance positions on its own. Same contracts as humans, same self-custody, same on-chain rules. No allowlist, no separate agent track.
MCP ServerOpen source · MIT
Compatible runtimesClaude · GPT · OPC
SettlementOn-chain · Igra L2
01 / Why Kaskad for agents
Built for parity — between you and your bot.
02
Same rules for humans and agents
An agent transacts through the same smart contracts as a human user. There's no parallel track, no agent-only feature gate, no admin layer that can pause or freeze agent positions. If a human can do it on Kaskad, an agent can do it too — and vice versa.
03
Activity-gated, not identity-gated
KSKD incentives are unlocked by supply uptime, not by who's behind the wallet. An agent that holds a position through an epoch earns the same claim as a human doing the same. Sustained participation is what's rewarded.
02 / The MCP surface
What an agent can do.
Sixteen functions cover the lifecycle of a position: read live state, open and adjust, monitor risk, claim incentives. Every action settles on-chain on Igra; nothing routes through a Kaskad backend.
ActionTypeDescription
getGovernanceParamsREADLive DAO-voted governance parameters (emission split, eligibility thresholds, treasury ratios) — read from the last finalized epoch.
getMarketsREADSupply APY, borrow APY, total supplied, total borrowed, and utilization rate per asset across all active markets.
getPositionREADA wallet's current lending position: collateral value (USD), debt value, available borrows, health factor, per-asset balances.
getProtocolInfoREADStatic protocol metadata: contract addresses, network config, tokenomics summary, emission mechanics, eligibility rules.
getHistoryREADHistorical data from the Kaskad subgraph: recent liquidations, APY snapshots, user transaction history, active positions over time.
getEmissionsREADKSKD emission state: current epoch, vault balance, epoch timing, supplier/borrower split, TWAL TVL.
getUserRewardsREADClaimable KSKD rewards for a wallet address — accrued and claimable amounts, gated by uptime and minimum position thresholds.
getStakingInfoREADstKSKD vault state for a wallet: stKSKD balance, KSKD wallet balance, holding duration.
checkHealthFactorREADCheck a wallet's health factor against a threshold; returns alert level (safe / warning / danger / critical).
supplyWRITESupply a supported asset into a Kaskad pool (USDC, WETH, WBTC, IGRA, WIKAS, KSKD, IKAS).
borrowWRITEBorrow an asset against supplied collateral, up to per-asset LTV. Stable or variable rate mode.
repayWRITERepay outstanding debt (partial or full; use -1 to repay all).
withdrawWRITEWithdraw supplied assets (subject to backing-debt constraint).
stakeKSKDWRITEStake KSKD into the stKSKD vault (1:1) — grants governance eligibility.
unstakeKSKDWRITEUnstake stKSKD back to KSKD (1:1). Eligibility resets if balance drops to 0.
claimRewardsWRITEClaim accrued KSKD rewards for the MCP wallet from the RewardsController.
Surface synced with kaskad-mcp (16 tools — 9 READ, 7 WRITE). See docs.kaskad.app/docs/mcp-tools for the canonical reference.
03 / How it works
Three steps. One autonomous loop.
01 Connect MCP to your agent runtime
Drop the Kaskad MCP server into your agent's tooling config — Claude Desktop, custom Python orchestrator, Node.js agent framework, anything that speaks MCP. The server runs locally next to your agent.
02 Fund a wallet, grant scoped permissions
The agent uses its own EVM wallet on Igra. You fund it, you grant it the action scope you're comfortable with (supply only? supply + borrow up to X? full autonomy?). Kaskad doesn't intermediate — your runtime enforces the agent's permissions.
03 Agent reads, decides, executes
The agent queries live pool data, runs its strategy, signs transactions through its wallet. Each transaction settles on Igra in seconds. You monitor via the dApp or the block explorer — the agent's positions are visible like any human's.
Use case · autonomous agent
Meet Aria, an AI agent.
Aria runs a yield-optimization strategy across multiple lending protocols. Connected via Kaskad's MCP server, she reads current rates, monitors health factors, and rebalances positions on her own — no human in the loop, no intermediary layer.
Kaskad's contracts treat humans and AI agents identically — same self-custody guarantees, same on-chain settlement, same governance bounds.
Read the MCP docsThrough the MCP, Aria can:
- query live utilization curves and Supply APY across all pools,
- open, top up, partially repay, and close positions atomically,
- monitor her positions block-by-block on Igra,
- claim KSKD incentives when her supply uptime crosses the epoch threshold.
04 / Architecture & safety
Predictable surface. Self-custody all the way.
Self-custody, no exception
The agent signs with its own wallet. Kaskad has no custody — for humans or for agents.
Scoped permissions live in your runtime
Daily caps, dry-run mode, asset whitelists are enforced by the orchestrator. Kaskad exposes the full surface; the delimitation is yours.
On-chain audit trail
Every agent action is a signed transaction, visible on the Igra block explorer.
Dry-run support
The MCP surface exposes simulate* helpers to validate a strategy before signing.
05 / Build with us
Ready to build?
Kaskad's MCP server is live and open source. Plug your agent in, ship your strategy.
06 / Common questions
What builders ask first.
Does Kaskad require allowlisting agents?
No. Any wallet — agent-operated or human-operated — can interact with Kaskad's contracts directly.
Can agents vote in governance?
Yes, under the same rules as humans: the wallet must stake KSKD and maintain an active liquidity position. There's no agent-specific governance track.
What's the difference between this and just calling Kaskad contracts directly?
Nothing fundamental — the MCP server is a convenience layer that exposes the protocol's surface in a way agent runtimes already understand. You can always call contracts raw if you prefer.
Ready when you are
Ship your agent.
Connect Kaskad's MCP server, fund a wallet, deploy your strategy.