Built for the agent era
Agents do the work. You approve what matters.
Nine open-source, local-first CLIs for contract operations — extract, draft, review, lint, compare, convert, sign, and track contracts from the command line. Each works standalone; together they cover the whole workflow. Built so an AI agent can run the operational work over MCP, while a human approves the gates that matter — starting with signing. Bring your own agent, or skip the wiring and run the ready-made contract-ops-agent (beta): one command, your model of choice, the whole suite in an enclosure.
Each box is a separate CLI; the arrows are the handoffs between them — files and JSON in standard formats. No proprietary glue, no shared database, no daemon — just standalone tools that compose.
Nine tools at a glance
| Tool | Stage | Reads → writes |
|---|---|---|
| extract · Python | Ingest | any contract → structured JSON |
| template-vault · Python | Store | vault → versioned template |
| draft · Node | Draft | template + params → filled draft |
| nda-review · Python | Review / negotiate | draft → verdict + redlines |
| contract-lint · Python | Lint gate | one contract → defect findings |
| compare · Node | Drift gate | two versions → pass / block verdict |
| docx2pdf · Node | Convert | .docx → .pdf |
| sign · Node | Sign | .pdf → signed .pdf + receipt |
| contract-vault · Python | Manage | signed contract → renewals + deadlines |
Every tool answers --catalog json for uniform agent discovery. Exit-code meanings differ per tool — see built for agents.
Try the NDA review live
Score an NDA against a house policy book and watch it flag every non-compliant clause, in a sandbox. No install. No sign-in.
Open the live demo →Two ways to use it
One command, bring your own model. contract-ops-agent is an enclosed
agent whose only tools are the suite — it drafts, reviews, compares, and tracks, and asks
before it writes a file or signs anything.
Every CLI answers --catalog json, and contract-ops-mcp
exposes all nine over MCP to any client (Claude Code, Cursor, Codex). Same guardrails — signing
stays behind a human.
Either way the split is the same: the agent does the operational work — drafting, scoring, amending, sending, tracking — and a human approves what legitimately needs one, starting with signing. The principles spell out what these tools refuse to do, and the comparison covers why nine small tools beat one big platform for an agent-first workflow.
Where to go from here
- 5-minute tour — run all nine zero-config demos, then chain them end to end. The fastest way in.
- The workflow — how the nine tools chain together, with copy-paste commands.
- Install — every install path on one page.
- MCP guide — drive the whole suite from an agent (Claude Code, Cursor, Codex).
- Comparison — how the suite stacks up against SaaS contract platforms.
- Principles — what these tools have in common and what they refuse to do.
- Changelog — what's new and what's coming next.
Or jump straight to any tool from the table above.