# Wordsmith Build — Production-grade legal AI for engineers using Claude Code and Cursor > MCP primitives for legal engineers. Real tracked-changes DOCX, playbook-driven review, human-in-the-loop escalation, audit-grade governance. Free signup for agents. No human form. Canonical URL: https://www.wordsmith.ai/build Free for your agent · Live in Claude Code ## Build production-grade legal AI The fastest way to become a legal engineer. You build the app. Wordsmith handles redlining, secure document sharing, and audit trails — the infrastructure that makes Claude enterprise-ready. [Start building ](#start)[Open in Claude Code ](#start) No signup form. Your agent provisions the workspace. Pay when you scale. ``` Use Wordsmith to read the Terms of Service at https://stripe.com/legal/ssa. Flag the five clauses an in-house lawyer would push back on. Cite section numbers and severity. Output a one-page memo I can forward to my GC. ``` ``` import { Wordsmith } from '@wordsmith/sdk'; const ws = new Wordsmith({ apiKey: process.env.WORDSMITH_API_KEY }); // Fetch a live ToS, extract risky clauses, output a one-page memo. const doc = await ws.ingest({ url: 'https://stripe.com/legal/ssa' }); const review = await ws.review({ document: doc.id, lens: 'in-house-counsel-pushback', limit: 5, }); const memo = await ws.draft({ template: 'one-page-risk-memo', source: review.id, }); return memo.url; ``` 3 primitives · 1 paste · A risk memo on any live ToS — no upload Capabilities ### All the tools your legal app needs. Every capability is one or more MCP primitives. Expand to see the building blocks. #### Store & search documents A versioned legal repository with provenance, semantic search, and audit-grade history. 2 primitives - ##### wordsmith.repository Versioned, queryable document store with provenance baked in. Claude has no memory across runs. We persist with audit-grade history. [View docs ](https://www.wordsmith.ai/docs/repository) - ##### wordsmith.ingest OCR, parse, and normalize PDFs, DOCX, and email threads into structured documents. Claude alone reads one file at a time. We hand you a typed corpus. [View docs ](https://www.wordsmith.ai/docs/ingest) #### Review & redline Playbook-driven review, clause-level diff, and real tracked-changes DOCX out the other end. 3 primitives - ##### wordsmith.review Playbook-driven review with versioned positions and severity scoring. Claude alone has no concept of your playbook. We anchor every comment to it. [View docs ](https://www.wordsmith.ai/docs/review) - ##### wordsmith.redline Native tracked-changes DOCX with real Word revision markup and sanitized metadata. Claude alone describes changes. We produce the actual file opposing counsel will open. [View docs ](https://www.wordsmith.ai/docs/redline) - ##### wordsmith.compare Semantic diff across versions with clause-level mapping. Claude alone produces a textual diff. We anchor to legal structure. [View docs ](https://www.wordsmith.ai/docs/compare) #### Drafting First drafts composed from your approved clause library and the counterparty profile. 1 primitive - ##### wordsmith.draft Generate first drafts from a template plus your clause library plus the counterparty profile. Claude alone hallucinates clauses. We compose from your approved library. [View docs ](https://www.wordsmith.ai/docs/draft) #### Bulk document analysis Schema-enforced extraction across thousands of contracts, with resumable jobs and continuous scans. 3 primitives - ##### wordsmith.extract Pull structured fields from any legal document with schema enforcement. Claude alone returns prose. We return validated JSON against your schema. [View docs ](https://www.wordsmith.ai/docs/extract) - ##### wordsmith.extract.bulk Run extraction across thousands of documents with rate-limited, resumable jobs. Claude alone cannot durably run 10,000 extractions and resume on failure. [View docs ](https://www.wordsmith.ai/docs/extract.bulk) - ##### wordsmith.scan Continuously scan your repository for regulatory or policy drift. Claude alone is a request/response system. We run continuously. [View docs ](https://www.wordsmith.ai/docs/scan) #### Translation Legal-grade translation via DeepL, with terminology locked to your bilingual glossary. 1 primitive - ##### wordsmith.translate Translate contracts and clauses through DeepL with your terminology mappings locked in. Claude alone has no contracted DeepL routing or your enforced bilingual term mappings. [View docs ](https://www.wordsmith.ai/docs/translate) #### Legal research Jurisdiction-aware research with citations grounded in primary sources. 1 primitive - ##### wordsmith.research Jurisdiction-aware legal research with citation grounding. Claude alone cannot verify citations against authoritative sources. [View docs ](https://www.wordsmith.ai/docs/research) #### Escalation & task management Pause the agent, route to a named reviewer, and track work across humans and agents. 2 primitives - ##### wordsmith.escalate Pause the agent and hand off to a named human reviewer with full context. Claude alone has no human-in-the-loop primitive. We make it first-class. [View docs ](https://www.wordsmith.ai/docs/escalate) - ##### wordsmith.tasks Assign, track, and deadline work across humans and agents in one queue. Claude alone has no persistent task store, deadlines, or assignment graph. [View docs ](https://www.wordsmith.ai/docs/tasks) #### Governance & audit Eval discipline against goldens and a regulator-grade trail of every agent action. 2 primitives - ##### wordsmith.evaluate Eval discipline against golden sets to catch prompt and model drift. Claude alone has no eval harness. We make prompt regressions visible. [View docs ](https://www.wordsmith.ai/docs/evaluate) - ##### wordsmith.audit Regulator-grade audit trail of every agent action, input, and output. Claude alone has no persistent trace. We log every primitive call defensibly. [View docs ](https://www.wordsmith.ai/docs/audit) [Read the full primitive reference ](https://www.wordsmith.ai/docs) Hello world ### Hello world. 60 seconds. Pick how you install. Paste the prompt. Your agent reads a live Terms of Service and returns a risk memo — no upload, no playbook, no dependencies. ``` claude mcp add wordsmith ``` ``` cursor mcp add wordsmith ``` ``` curl -fsSL https://www.wordsmith.ai/install.sh | sh wordsmith login wordsmith init ``` ``` npx @wordsmith/cli init # no install, just runs ``` ``` docker run --rm -it \ -e WORDSMITH_API_KEY \ ghcr.io/wordsmith/cli init ``` ``` code --install-extension wordsmith.wordsmith # then: cmd-shift-p → "Wordsmith: Sign in" ``` Try it — live ToS risk memo ``` Use Wordsmith to read the Terms of Service at https://www.wordsmith.ai/terms-of-service. Surface the top five clauses an in-house lawyer should review, with section numbers and severity. Return a markdown summary. ``` Pre-built apps ### Install a working legal app in one prompt. Six production-shaped apps, built on Wordsmith primitives. Paste the install prompt into Claude Code; your agent provisions the workspace, the playbook, and the escalation routes. - #### NDA Reviewer Redline incoming NDAs against your playbook. Real tracked-changes DOCX out. - wordsmith.ingest - wordsmith.review - wordsmith.redline - wordsmith.escalate [Install in Claude Code ](#start) - #### Vendor Renewal Watchdog Watch every vendor contract for auto-renewal windows. Alert 60 days out. - wordsmith.ingest - wordsmith.extract.bulk - wordsmith.repository - wordsmith.scan [Install in Claude Code ](#start) - #### Contract Change Detector Diff every new contract version against the last. Surface material changes only. - wordsmith.compare - wordsmith.review - wordsmith.audit [Install in Claude Code ](#start) - #### Employment Agreement Bot Draft jurisdiction-aware employment agreements from your clause library. - wordsmith.draft - wordsmith.research - wordsmith.review [Install in Claude Code ](#start) - #### RFP Playbook Scorer Score inbound RFPs against your win/no-bid playbook. Flag dealbreakers. - wordsmith.ingest - wordsmith.extract - wordsmith.review - wordsmith.escalate [Install in Claude Code ](#start) - #### DPA Auto-Reviewer Review inbound DPAs against your data-processing standards. Auto-redline. - wordsmith.ingest - wordsmith.review - wordsmith.redline - wordsmith.audit [Install in Claude Code ](#start) FAQ ### What lawyers ask before they let an agent touch a contract. The questions GCs, security teams, and procurement ask first. Click to expand. - Is the audit trail admissible? Every primitive call logs inputs, outputs, prompts, model version, and operator identity to an immutable trace. Exportable in formats your regulator and your court accept. - Does the agent respect ethical walls? Workspaces enforce matter-level and client-level access controls. An agent acting in one matter cannot read documents from another. Ethical wall configurations are auditable. - Can my GC stop an agent mid-flow? Yes. Wordsmith.escalate hands off to a named reviewer before any irreversible action. The agent waits. The GC approves, edits, or cancels. - Is attorney-client privilege preserved? Wordsmith is configurable as a service provider under common privilege frameworks. We sign BAAs, support ZDR for sensitive workspaces, and document the privilege posture for your auditors. - Can I prove what the agent did six months ago? Every action is captured with full provenance — primitive, input, output, model, prompt, operator, timestamp. Retention is configurable. The trail survives model upgrades. - What would I be deploying? Nothing. Wordsmith is a managed SaaS product. There is no on-prem component to install and nothing to patch. Your engineers integrate via MCP or the SDK. - What would I be approving? Outbound network access to *.wordsmith.com, Wordsmith’s connectors to your document sources (configurable), and any LLM provider routes you elect. All ingress is denied by default. [Full trust documentation at trust.wordsmith.com ](https://trust.wordsmith.com) Pricing ### Free signup. Pay when you scale. Three tiers. One billing surface. No per-seat tax on the legal engineers building on top. Free Start free Sign up your agent. No card needed. - Agent self-signup, no card - Evaluation usage on the house - All primitives unlocked - Card prompt at evaluation limit [Start free ](#start) Most popular Production Usage-based Pay only for what your agents ship. - All primitives, production-grade - Audit-grade trace + retention controls - Spend caps with hard pause - SSO, RBAC, SCIM - BAA + ZDR available - Email + community support [Get started ](#start) Enterprise Wordsmith Black Bespoke deployment. Procurement-grade. - Dedicated infrastructure - Custom retention + residency - Named deployment engineer - Procurement + DPA support [Talk to us ](https://www.wordsmith.ai/contact) Spend capProduction usage has a default $200 monthly budget. Set your own spend caps. Hard pause at 100%. No silent overage. [See full pricing ](https://www.wordsmith.ai/build/pricing)[Talk to sales ](https://www.wordsmith.ai/contact) Trusted by in-house legal teams at ![Canva logo](https://cdn.sanity.io/images/fzjkft9j/production/29cf002826a3547d7bb6371941970ac416c74a25-48x15.svg) ![Remote logo](https://cdn.sanity.io/images/fzjkft9j/production/cd7c00e8e3efafda58a3d7c2951ccdeb1041ec48-64x14.svg) ![Trustpilot logo](https://cdn.sanity.io/images/fzjkft9j/production/9c8919a948f09294954dee08f20bdb43b73c63ca-71x17.svg) ![Nikon logo](https://cdn.sanity.io/images/fzjkft9j/production/a0460fde77ac05f975ab5c720917318e1a07eae6-63x16.svg) ![Deliveroo logo](https://cdn.sanity.io/images/fzjkft9j/production/c000d94105d3b4ef9ca691283ac31a452ba65d0d-73x22.svg) ![Financial Times logo](https://cdn.sanity.io/images/fzjkft9j/production/06b6ce83e9d56a59d0e17a73610647dc1a0c33c3-130x11.svg) ![Skyscanner logo](https://cdn.sanity.io/images/fzjkft9j/production/d3d693982fd93d4ec81ba08328013531a7a7edd7-64x31.svg) ![BT logo](https://cdn.sanity.io/images/fzjkft9j/production/788fddbfd69c38e3af68646406e38f0261e72a32-31x31.svg) ![Trip.com logo](https://cdn.sanity.io/images/fzjkft9j/production/935eae5ed13a100d5d7a3c8f5c0018349fa03b66-90x22.svg) > “Wordsmith made me the ultimate legal engineer.” Head of Legal · Belron Manifesto ### We believe legal work is moving from humans-with-tools to agents-with-oversight. The platform that serves agents safely — with provenance and the governance lawyers can defend — will define the next decade of legal AI. Wordsmith is built for that world. Vibe-code the fun stuff. We handle production. Ross McNairn · Founder & CEO [Start building ](#start) No human signup form. Your agent provisions the workspace.