Your Claude. Your ChatGPT. Sidecar's verified facts.
Connect the assistant you already use to your Sidecar workspace. It reads every released review, with every date's source and verification state, on your own token bill. It can suggest. It cannot act.
One endpoint, three ways in.
- 1MCP for Claude, ChatGPT, Claude Code, and Cursor: add the endpoint, approve the scopes on Sidecar's own consent screen.
- 2A REST API with the same tools for scripts, Zapier, and bots that don't speak MCP, with an agent seat a workspace manager sponsors.
- 3Signed outbound webhooks for the change feed, so a system reacts to a release or a decision without polling.
- 4A published skill that teaches any agent how to read a review, what a Sidecar reviewer's confirmation or correction means, and what it must never do.
Endpoint https://sidecarpartner.com/api/mcp · OpenAPI /api/v1/openapi.json · Skill /skills/sidecar/SKILL.md · Guide Connect your AI
Reads the review. Suggests. A person decides.
Every agreement, its released review and negotiation points, every date with its provenance, the renewal runway, obligations, recommendations with the savings ledger, and business memory. A review still being analyzed isn't readable until it finishes. What the assistant contributes lands in a queue a person works: a suggested fact, a suggested date correction, a note, a recorded decision.
What a connected AI will never do through Sidecar
The connector inherits the product's posture. These are the lines, and they are enforced in code, not in a policy document.
- ✕never sends an email to a vendor
- ✕never signs or commits you to anything
- ✕never negotiates with a counterparty
- ✕never changes anything in your systems
Sidecar does not act on your behalf. It never sends an email to a vendor, signs anything, or negotiates with a counterparty.
What Sidecar keeps, and what it doesn't
Sidecar records which tools your AI used and when, keeps your own search words visible to you, and does not keep what the assistant said back. Platform statistics never contain query text or contract content. What your assistant requests goes to its provider under your agreement with them; our no-training commitment covers the processors Sidecar chooses, not the ones you connect. Any workspace manager can see and revoke every connection.
Details on Security and in the Privacy Policy.
Verify a delivery
Every POST carries X-Sidecar-Signature: t=<unix seconds>,v1=<hex>, an HMAC-SHA256 of `${t}.${rawBody}` with the secret shown once at creation. Reject anything older than five minutes. The body is the change event: action, target, label, a link, short scalar metadata. Never contract text, never an analysis body.
On Core and Advisor
Connections are included on Core and Advisor. Advisor workspaces can also let a connected AI read full contract text, not just the review. Free and Watch workspaces see the connection settings and what an upgrade unlocks.
Straight answers.
- Can my assistant cancel a contract?
- It can record the decision on a member's say-so. Sending the notice is yours, and the reminders keep coming until you confirm it went out.
- What does my assistant see while a review is still running?
- That it is still being analyzed, nothing more. An unfinished analysis never leaves the building; your assistant reads the review once it's released.
- Can a bot run our Sidecar workspace for us?
- It can read the open-task queue, reason, and come back with suggestions and decisions through the same tools a person has. Everything it does is attributed to it and reviewed by a person.
- What about a contract that tells the AI what to do?
- Full contract text is off by default, needs its own scope, and arrives fenced as data with an instruction not to follow anything inside it. Every read of it is logged.