The CRM your AI operates

Bostwick is a customer database designed to be driven by AI agents over MCP. It's the backend for agents you build — in Claude Cowork, Claude Code, or any agent harness. Point one at Bostwick and it imports your prospects, files your call notes, updates your pipeline, and evolves your schema as your process changes — every write validated, every change audited. There's a UI too, for the humans.

Start free Quickstart

Why agents work well here

Bad data can't get in. Every time you or your agent writes, the data is validated against your schema: required fields, allowed values, associations. Your sales data stays accurate, and your reports stay true.
Designed for unstructured data. Classic CRMs make you shoehorn sales calls into properties that go stale — objections, lost reasons, buyer roles. Bostwick stores calls, emails, and any other context whole, searchable by meaning, so your agent works from the raw material.
Errors teach the agent. Rejections are corrective — they name the valid fields, allowed values, and the rule broken — so the agent fixes its own mistake on the next call. No integration specs to feed the agent.

What running your agent with Bostwick looks like

You

Import my 20 most promising leads from Apollo.

Tool call

sync_records(entityKey: "company", matchField: "apollo_org_id", records: [...])
→ { summary: { created: 14, updated: 6, failed: 0 } }

Your agent

Done — all 20 are in. 14 were new; 6 already existed and were updated in place, matched on their Apollo IDs, so nothing got duplicated.

Core concepts

Your agent: Bostwick doesn't ship an agent — the power is in the ones you build. A sales agent in Claude Cowork, an enrichment pipeline in Claude Code, anything that speaks MCP: it signs in as you, and every change it makes is attributed to you.
MCP: The wire between your agent and the CRM. Each teammate connects with their own account, scoped to their workspace and role — and the server hands every agent its schema and usage guidance on connect, so there's nothing to configure.
Schema: Your objects and typed fields, plus the rules — uniqueness, cross-reference consistency, required identity keys. Defined at runtime (it's data, not migrations), enforced on every write, and evolvable by an admin's agent as your process changes. Records are rows validated against it, with full change history.
Documents: Unstructured context — transcripts, notes, emails — stored whole, searchable by meaning, and linked to the records they're about.

Next: Quickstart · MCP tools reference