Claude gives lawyers the AI.
Verdict makes it defensible.
On May 12, 2026 Anthropic released Claude for Legal — 12 practice-area plugins, 80+ specialist agents, 20+ MCP connectors. Source-attributed citations and privilege guardrails ship with it. Cryptographic sealing, chain of custody, and FRE 902(14) authentication packets do not. That is the gap Verdict was built to fill.
“Every output from these plugins is a draft for attorney review — not legal advice, not a legal conclusion, not a substitute for a lawyer.”
— anthropics/claude-for-legal, README. The attorney signs. The record is what survives the signature.
20+ connectors
DocuSign, Ironclad, iManage, NetDocuments, LexisNexis, Thomson Reuters CoCounsel, Box, Everlaw, Trellis, Aurora, Definely, Solve, TopCounsel, CourtListener, Linear, Jira, Asana, Lawve, Courtroom5, Descrybe, Slack, Google Drive — all wired into Claude for Legal at launch.
12 plugins · 80 agents
Commercial, corporate, employment, privacy, product, regulatory, AI-governance, IP, litigation, legal-clinic, law-student, builder-hub. Apache 2.0 — usable commercially, modifiable, redistributable.
Forensic layer
Anthropic shipped accuracy guardrails (citation attribution, conservative privilege defaults). It did not ship cryptographic sealing, chain of custody, transparency-log anchoring, or FRE 902(14) authentication packets. That is Verdict.
Sequential.
Not competitive.
Claude for Legal
Drafts contracts, reviews clauses, researches authority, triages matters, summarizes documents, and coordinates legal workflows across the systems lawyers already use.
Verdict
Captures exactly what happened, hashes the relevant inputs and outputs, binds attorney review state, links the event into a chain of custody, and renders the record for regulators, insurers, and courts.
Every legal AI action
needs a record.
| Workflow | Claude legal layer | Verdict evidence layer |
|---|---|---|
| Contract review | Reviews NDA or vendor agreement against the firm playbook. | Seals prompt hash, source document ids, model version, clause deltas, reviewer disposition, and final export hash. |
| Brief drafting | Drafts argument, summarizes authority, and prepares first-pass motion structure. | Seals cited authority ids, prompt packet hash, generated work product hash, attorney revisions, and FRE 902(14)-style custodian packet. |
| Docket watcher | Monitors filings, deadlines, court notices, and matter tasks. | Seals the triggering event, deadline calculation, alert delivery, escalation state, and any missed or overridden action. |
| AI governance review | Classifies proposed AI use, reviews vendor terms, and drafts policy language. | Seals the classification path, policy bundle version, reviewer approval, retention rule, and board-ready audit export. |
The MCP is live
as a sandbox surface.
The first public surface is intentionally narrow: it lets Claude create a sandbox evidence record, score provenance completeness, and draft an authentication packet. Production tenants add OAuth, retention, HSM signing, Rekor anchoring, and matter-scoped access controls.
# Claude Code (stdio) — local install
claude mcp add verdict -- npx -y @verdict-systems/mcp
# Claude Cowork / claude.ai — remote endpoint
# Settings → Connectors → Add custom
https://verdict.systems/api/mcpVerdict speaks JSON-RPC 2.0 / MCP protocol 2025-06-18. Apache 2.0. The remote endpoint is the same surface every Claude for Legal plugin can reference from its .mcp.json. Full install guide →
verdict_create_evidence_recordCreates a deterministic sandbox SER for a legal AI output with payload hash, Merkle root, review state, citations, and chain-of-custody fields.
verdict_score_evidence_readinessScores whether an AI legal output has enough provenance for audit, discovery, outside counsel guidelines, and authentication preparation.
verdict_export_fre902_certificateDrafts a custodian-review packet from a sealed record so counsel can prepare a qualified certification.
POST https://verdict.systems/api/mcp
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "verdict_create_evidence_record",
"arguments": {
"matter_id": "M-100",
"workflow": "brief_drafting",
"input_summary": "Draft motion from approved authority packet.",
"output_summary": "Generated argument outline with cited authorities.",
"human_review_status": "pending",
"model_provider": "Anthropic",
"model": "Claude Opus 4.7"
}
}
}Directory connector first.
Plugin package second.
Remote MCP live
/api/mcp is JSON-RPC 2.0 compliant and serves the three Verdict tools at MCP protocol 2025-06-18. Production tenants add OAuth, retention, HSM-bound Ed25519 signing, and Sigstore Rekor anchoring.
Upstream PR
Submit @verdict-systems/mcp to anthropics/claude-for-legal as a connector across commercial-legal, corporate-legal, litigation-legal, ai-governance-legal, ip-legal, and regulatory-legal — the six plugins where AI-output authentication is most exposed to FRE 707 and state-bar disclosure rules.
Skill packaging
Author the Verdict sealing skill as an installable plugin under legal-builder-hub. Every contract reviewed, brief drafted, docket alert, and AI-governance triage becomes one tool call away from sealed.
Anthropic partnership
Position alongside Thomson Reuters CoCounsel as the independent evidence layer. CoCounsel: workflow + research. Verdict: workflow + record. Neither competes with Anthropic's surface — both extend it.
Status: independent integration strategy. Verdict is not presented here as an Anthropic partner or endorsed connector until the directory and partnership process makes that true.
The legal AI boom creates work product.
The evidence layer creates defensibility.