Every legal AI tool promises to transform how lawyers work. Most of them are general-purpose chatbots with a legal marketing wrapper. Harvey is not that. It is the first platform I have seen that is genuinely built around how legal professionals interact with documents, and the adoption numbers reflect it: half of the AmLaw 100 now runs on Harvey, with over 1,000 customers across 60 countries.
The question is not whether Harvey is capable. It clearly is. The question is whether the cost, which starts at roughly $288,000 per year before you even negotiate, makes sense for your firm. This review breaks down what Harvey actually does, where it excels, where it falls short, and who should be paying attention.
- Vault lets you analyse up to 100,000 documents per project with 97% field extraction accuracy, turning mass document review from weeks into hours
- Workflow Builder has enabled firms to create 15,000+ custom legal workflows without writing code
- Multi-model routing selects between Claude Opus 4.6, Sonnet 4.5, GPT-5.1, and GPT-5.2 depending on the task
- Native MCP support as both client and server means Harvey can integrate into existing firm toolchains
- Pricing is prohibitive for most firms: approximately $1,200 per lawyer per month with 20-seat minimums
How Harvey handles context
Legal work is document work. A merger review might involve tens of thousands of contracts. A regulatory compliance check might require cross-referencing filings across multiple jurisdictions. The fundamental challenge is not whether an AI can read a single document. It is whether the AI can reason across an entire corpus of documents while understanding the specific legal context that makes each one relevant.
Harvey's approach centres on Vault, a collaborative document analysis environment that can hold up to 100,000 documents per project. You upload your document set, and Vault creates a structured, searchable interface that looks more like a sophisticated spreadsheet than a chatbot window. The AI extracts fields, identifies clauses, and organises information across the full document corpus. The extraction accuracy sits at 97%, which is high enough to be genuinely useful but not high enough to eliminate the need for human review on critical matters.
- Vault and legal context
Vault is Harvey's document intelligence layer. Unlike general-purpose AI tools that process documents one at a time, Vault lets legal teams load entire deal rooms, contract portfolios, or regulatory filing sets and reason across them simultaneously. The 97% extraction accuracy means roughly 3 in 100 fields will need manual correction, a meaningful improvement over manual review but not a replacement for it.
What makes this more than a document management system is Workflow Builder. Firms can configure multi-step legal processes without code. Lease analysis, M&A due diligence, contract comparison, regulatory review. Customers have created over 15,000 custom workflows, which tells you something about how firms are actually using the tool. These are not one-off experiments. They are embedded operational processes.
Harvey also operates as both an MCP client and an MCP server. As a client, it can call external services and pull in data from other platforms. As a server, it exposes its own capabilities to other AI tools in your firm's stack. This bidirectional MCP support is rare and genuinely useful for large organisations that need Harvey's capabilities to integrate with existing infrastructure rather than replacing it entirely. The platform routes between multiple AI models, including Claude Opus 4.6, Claude Sonnet 4.5, GPT-5.1, and GPT-5.2, automatically selecting the best model for each task.
What Harvey gets right
The most impressive thing about Harvey is not any single feature. It is the fact that the entire platform is built around how lawyers actually think about problems. General-purpose AI tools treat every document the same way. Harvey understands that a lease agreement, a merger filing, and a regulatory submission are fundamentally different kinds of documents that require different kinds of analysis.
Vault is the standout. Being able to load 100,000 documents into a single project and run structured extraction across all of them is a capability that simply does not exist in general-purpose AI tools. A due diligence review that would take a team of associates weeks of manual work can be reduced to hours of structured AI-assisted review. The 97% accuracy rate on field extraction means the AI handles the bulk of the mechanical work while lawyers focus on the judgement calls that actually require their expertise.
The LexisNexis strategic alliance adds another dimension. Harvey integrates primary law content and Shepard's Citations directly into its platform, which means lawyers can verify citations and check case validity without leaving the Harvey environment. This is the kind of integration that makes a tool feel like part of your workflow rather than an addition to it.
Repsol's deployment data gives a concrete picture of the productivity impact. After rolling out Harvey to their in-house legal team, they reported 4 to 6 hours saved per lawyer per week at 96% adoption. Those numbers suggest that lawyers are not just experimenting with the tool. They are using it as a core part of their daily workflow. HSBC's strategic partnership, announced in March 2026, signals that Harvey is moving beyond law firms into global financial institutions.
Where Harvey falls short
The pricing is the most significant barrier. At approximately $1,200 per lawyer per month with 20-seat minimums and 12-month commitments, the annual floor is $288,000. Add the LexisNexis bundle at $400 to $600 per lawyer per year and the total cost climbs further. This is enterprise pricing in the truest sense. Solo practitioners, boutique firms, and most mid-size practices are priced out entirely. There is no self-serve signup, no free trial, and no pathway for smaller organisations to test the platform before committing.
Harvey can produce incorrect citations. This is not unique to Harvey. Every large language model can hallucinate legal references. But the stakes in legal work are higher than in most professional domains. An incorrect citation in a court filing has real consequences. The LexisNexis integration helps with verification, but it does not eliminate the need for careful fact-checking. Lawyers using Harvey still need to treat its outputs as drafts that require professional review, not finished work product.
The learning curve is steep. Adopting Harvey is not like signing up for a new SaaS tool. It requires significant change management: training lawyers to use Vault effectively, building custom workflows in Workflow Builder, integrating with existing document management systems. Firms that succeed with Harvey tend to invest in dedicated implementation teams. Firms that treat it as a plug-and-play solution tend to underutilise it.
Jurisdiction coverage also has gaps. Despite access to over 100 legal data sources, Harvey's knowledge is uneven across regions. Firms doing multi-jurisdictional work should verify coverage for their specific practice areas before committing.
Feature analysis
| Feature | Harvey |
|---|---|
| Context Persistence | Full support |
| Context Portability | Not supported |
| MCP Support | Full support |
| Cross-Platform Compatibility | Partial support |
| Data Sovereignty | Full support |
| Knowledge Management | Full support |
| Enterprise Readiness | Full support |
| Agentic Capabilities | Full support |
| Domain Specialisation | Full support |
Our take
Harvey is the most capable AI tool built specifically for legal professionals. Vault's ability to process 100,000 documents at 97% extraction accuracy, combined with Workflow Builder's 15,000+ custom legal workflows and multi-model AI routing, makes it a genuine productivity multiplier. The data backs this up: Repsol reported 4 to 6 hours saved per lawyer per week at 96% adoption. The barrier is cost. At roughly $1,200 per lawyer per month with 20-seat minimums, Harvey is designed for AmLaw 100 firms and in-house legal teams at global enterprises. HSBC's strategic partnership confirms the trajectory. Solo practitioners and small firms need to look elsewhere until Harvey introduces a more accessible tier.
Who Harvey is for
Harvey is built for large law firms and in-house legal departments at global enterprises. If your firm handles high-volume document review, complex multi-jurisdictional matters, or large-scale M&A due diligence, Harvey's capabilities are difficult to replicate with general-purpose AI tools. The 50% AmLaw 100 adoption rate tells you where the product-market fit sits.
It is not for solo practitioners, small firms, or legal professionals who primarily need a research assistant or drafting tool. The pricing structure, implementation requirements, and enterprise-only sales process all point to a product designed for organisations with significant legal operations budgets. If your annual technology spend per lawyer is measured in hundreds rather than thousands, Harvey is not currently built for you. For firms in that position, a combination of Claude or ChatGPT with legal-specific prompt engineering and a Westlaw or LexisNexis subscription may deliver 70% of the value at a fraction of the cost.