Harvey Says It Built an MCP Policy Engine to Cut Down Agent Tool Risks
Blockchain.News reported that Harvey detailed an MCP Policy Engine adding runtime controls for tool access and agent actions.
Why it matters
The relevance is narrower than the framing suggests. MCP Policy Engine work is a vendor's account of its own product security controls, reported by a single outlet, with no independent testing or evidence that any customer was affected. The stated customer count and country reach are the only scale figures in evidence. No token, price or market effect is involved.
Key facts
Harvey announced development of a Model Context Protocol Policy Engine for its platform.
Reported by Blockchain.News
The engine was described in a Harvey blog post published on 7 October 2026.
Reported by Blockchain.News
The policy engine adds runtime controls governing tool access, information flow and agent actions.
Reported by Blockchain.News
Harvey said the engine responds to prompt injection, tool poisoning and rug-pull attacks, the last covering tool definitions altered after approval.
Reported by Blockchain.News
Harvey said its mitigations include least privilege restrictions, complete mediation of tool actions and secure information flow controls.
Reported by Blockchain.News
Harvey said it reviews partner MCP integrations and enforces security policies at runtime.
Reported by Blockchain.News
Harvey reported serving more than 3,000 customers across over 70 countries.
3,000 customersReported by Blockchain.News
The engine includes a Tool Pinner that tracks and flags tool definitions, according to the report.
Reported by Blockchain.News
What happened
Blockchain.News reported on8 October 2026 that Harvey, an AI company for legal and professional services workflows, has developed a Model Context Protocol (MCP) Policy Engine. The outlet said the work follows a blog post Harvey published on 7 October 2026, and that the engine adds runtime controls over tool access, information flow and agent actions. Harvey wrote that MCP widens the attack surface for AI agents even as it extends what those agents can do, and named prompt injection, tool poisoning and rug-pull attacks — malicious tool definitions or post-approval changes — as threats the controls are meant to address. The company said it applies least privilege restrictions, complete mediation of tool actions and secure information flow controls, and reviews partner MCP integrations by assessing tool capabilities and risks before enforcing policies at runtime. Harvey also cited a user base of more than 3,000 customers across over 70 countries. Only Blockchain.News has reported on this, so the engine's existence rests on Harvey's own description and has not been independently confirmed or tested.
How the story developed
- Official statement
Harvey publishes blog post on its MCP Policy Engine
The post describes runtime controls for tool access, information flow and agent actions, and names prompt injection, tool poisoning and rug-pull attacks as the threats addressed.
- First report
First report by Blockchain.News
Harvey Develops MCP Policy Engine to Enhance AI Security
- Update
Blockchain.News reports on the MCP Policy Engine
The outlet described the engine's controls, Harvey's partner integration reviews and its stated customer base.
Affected entities
Original sources
| Publisher | Report | Role | Published |
|---|---|---|---|
| Blockchain.News Crypto media | Harvey Develops MCP Policy Engine to Enhance AI Security | Original |