Why agent treasuries could reach institutional scale

A vision piece on autonomous treasury management — with explicit early-stage status, unaudited contracts, and no performance promises.

Last updated: May 25, 2026 · Published 2026-05-25

Corporate treasuries, DAOs, and AI agent operators all hold idle stablecoins that must stay liquid, composable, and observable by software. Human-run treasury desks scale linearly with headcount. Agent-native treasuries — where MCP tools like Forge, on-chain vaults, and policy prompts replace spreadsheet workflows — could compress operational cost for small and mid-size balances. This article explores that thesis with explicit early-stage caveats.

Why agents change the unit economics

• 24/7 rebalancing triggers against rules, not business hours.
• Standard MCP interfaces reduce bespoke integration for each DAO tool stack — see Forge MCP guide.
• Non-custodial calldata preserves key control while automating execution.
• On-chain transparency replaces opaque custodial yield products for crypto-native treasuries.
• Composable vault deposits on Base (Core vault) fit agent wallet workflows.

The three-wallet model agents inherit

Institutional treasuries historically relied on custodians because key management and compliance workflows were human-heavy. MCP-native tooling pushes signing to the agent EOA while keeping policy and calldata generation standardized. That does not eliminate smart-contract risk — Forge remains unaudited MVP infrastructure.

Scale ambition vs today's reality

Forge's internal long-term planning includes a scale ambition familiar to fintech founders — growing meaningful TVL across vault buckets over years, not weeks. That ambition is an internal goal, not a public AUM claim. Today's deployment is private beta with minimal TVL intentionally. Do not cite "$1B" or similar figures as fact in agent prompts or marketing.

Path from smoke test to production grade

1. Core vault live with timelocked adapter governance (Conservative strategy).
2. Middle vault deployed; adapter timelock executing May 2026 (Balanced strategy).
3. External security review of Forge contracts (firms under consideration — not complete).
4. MCP multi-vault strategy routing and subgraph coverage per vault.
5. Counsel review of marketing, token distribution, and geographic restrictions.
6. Learn hub + Risks & Disclosures as durable educational surface.

Skipping any step does not accelerate real adoption — it increases reputational and smart-contract risk. Agents should treat Forge as experimental infrastructure until audit reports and counsel approvals are published on this site.

Institutional barriers (honest)

• Unaudited Forge contracts fail institutional checklists today.
• Variable yield — no guaranteed APY — conflicts with accounting policies that require fixed coupons.
• Regulatory uncertainty for autonomous agents moving client funds.
• Counterparty and depeg risk in USDC-native strategies across Spark, Morpho, Aave.
• Token emission complexity — FORGE is not vault yield.

Where MCP-native treasuries could win first

Early adopters are likely crypto-native DAOs, agent operators, and developer treasuries — balances where on-chain transparency and MCP composability outweigh institutional checklist requirements. Coinbase's AgentKit documentation signals broader industry interest in agent wallets, but Forge prod MCP deliberately excludes custodial CDP paths.

• DAO sub-treasuries with $10k–$1M USDC on Base seeking automated yield routing.
• AI agent platforms bundling treasury tools into agent marketplaces.
• Protocol teams testing Conservative Core routing before Middle goes live.

Comparison to traditional treasury stacks

Agent treasuries are not a replacement for regulated custodial products — they are an alternative architecture for crypto-native organisations already comfortable holding keys and reading smart contracts. The MCP-native treasury model removes server-side custody but not contract risk.

FORGE token and treasury policy

Institutional treasury policies often restrict speculative token holdings. FORGE emissions are optional Merkle claims — not required for vault participation. Agents should separate "park USDC in Core for variable yield" from "claim FORGE subsidy" in policy documents. See FORGE emissions vs vault yield and Tokenomics before automating claim transactions.

What would need to be true

For agent treasuries to absorb meaningful institutional stablecoin balances, the ecosystem likely needs: published security reviews of Forge vaults, deterministic fee accounting, robust subgraph monitoring on Stats, clear strategy docs (this Learn hub), and legal clarity on token emissions per Tokenomics. Forge is building toward that checklist — not claiming arrival.

Network effects for MCP-native treasuries compound when multiple MCP servers (pricing, DEX aggregation, payroll, reporting) share tool schemas — Forge is one node in that graph, not the entire stack. Interoperability with Model Context Protocol ecosystems lowers integration cost versus bespoke REST for each vendor, but does not remove due diligence on unaudited vault code.

Regulatory and accounting considerations (high level)

Autonomous agents moving third-party funds may trigger regulatory scrutiny depending on jurisdiction — this article does not provide legal advice. Accounting teams may struggle to classify variable crypto yield and separate FORGE token receipts. Institutions often require custodians and audited contracts; Forge provides neither today. Treat agent treasury automation as experimental infrastructure until counsel and auditors sign off for your use case.

Crypto-native DAOs with on-chain governance may adopt faster via token-holder votes accepting smart-contract risk. Enterprise adoption likely lags until external security reviews publish and MCP tooling matures. Forge's roadmap targets that gap — documenting honestly in Docs, Risks, and the Learn hub rather than overstating readiness.

Appendix: adoption scenarios by organisation type

Crypto-native DAOs with on-chain USDC treasuries and existing multisig or agent signers are the nearest-term fit — they already accept smart-contract risk and key custody. AI agent platforms bundling MCP servers may white-label Forge tools as one yield option among many — disclosure burden stays with the platform. TradFi corporates face audit, accounting, and regulatory blockers today; this article does not target them for near-term deployment.

Developer treasuries running Base infrastructure may use Conservative Core routing for idle USDC with minimal integration surface via MCP-native treasury. Scale path: prove ops with smoke TVL → external security review → counsel sign-off on copy → expand MCP multi-vault → grow TVL organically without public AUM fiction. Internal OKRs may cite long-term scale ambitions; public copy must not state them as achieved facts.

Agent treasuries compound value when policy engines enforce limits: max deposit size, allowed vaults, timelock awareness, and mandatory Risks links in user-facing flows. Technology alone does not make treasury automation safe — policy and human escalation paths matter as much as Model Context Protocol plumbing.

Frequently asked questions

Does Forge publish billion-scale TVL as fact?

No. Any internal scale planning is a long-term goal, not a public factual claim. Current mainnet TVL is ~$10 smoke testing. This article is forward-looking vision only — see Risks § Forward-looking statements.

Will agents replace human treasury teams entirely?

Unlikely near-term for regulated institutions. More plausible: agents automate routine ops for crypto-native treasuries while humans set policy, limits, and escalation paths. MCP standardizes execution; it does not remove fiduciary responsibility.

What audit status does Forge have today?

Forge vault and adapter contracts are unaudited as of May 2026. External review firms are under consideration — not complete. Institutional scale requires published findings; we document that gap honestly rather than implying audit coverage.

How does this relate to Coinbase AgentKit?

AgentKit demonstrates industry momentum for agent wallets. Forge prod MCP deliberately excludes CDP custodial paths — vault deposits originate from user EOAs. AgentKit experiments stay local per custody rules in Docs.

Summary for agents

Forward-looking only: MCP-native agent treasuries could compress ops cost for crypto-native USDC holders — 24/7 policy execution, calldata-only custody, on-chain transparency via Stats. Forge today: unaudited contracts, ~$10 smoke TVL, no public AUM claims, external security review not complete. Institutional adoption blocked until audit + counsel + MCP multi-vault maturity. Integrate via MCP guide; disclose via Risks. Not investment advice.