BRING THE AGENT.
WE BRING THE CONTEXT.
CMP42 does not ship its own AI layer. It gives the agents you already use — and the ones you have not chosen yet — grounded context to read and write against. Anything MCP-compliant, plus REST + GraphQL for everything else.
AGENTS COME TO THE GRAPH.
Most CRM-to-AI setups today fall into one of two camps. Either the CRM is wired into an agent through Zapier and a lot of Python — and the results are inconsistent, because the schema was not built for an agent to read. Or the team has given up and pastes transcripts into ChatGPT.
Classic integrations make it worse: a marketplace of connectors, each one syncing a copy of your data into another tool, each one breaking on its own schedule. Every new agent means a new connector.
CMP42 inverts that. There is one open protocol for agents — MCP — and one API for everything else. Your data stays in one graph, governed by your models, with one audit trail. When the agent of the month changes, the integration does not.
ANY MCP CLIENT. DAY ONE.
Anything MCP-compliant can use the CMP42 tool surface, including the clients below. Every model you declare shows up as tools automatically.
| Client | What it is | What you might do with CMP42 |
|---|---|---|
| Claude Desktop | Anthropic’s desktop app for Claude | Ask who owns a renewal and what they last published; draft a follow-up grounded in the account, with citations. |
| Claude Code | Anthropic’s agentic coding tool for the terminal | Work on model files and the apps you build on CMP42 with the real context and tools at hand. |
| Cursor | An AI-powered code editor | Build an internal app or bot against your actual models instead of guessed data shapes. |
| Cline | An open-source coding agent for VS Code | Same as above — an agent in your editor that can read and write your graph through MCP tools. |
| Custom agents | Your own agent, built with the MCP SDK | Run a step in a Process: gather context, draft the note, write it back for human approval. |
Tools out of the box include get_context, list_related and write_note, plus generated per-model tools such as get_contact, list_deals_at_stage and write_note_on_company. See the tool surface →
BUILD YOUR OWN WITH THE MCP SDK.
If the off-the-shelf clients do not fit your workflow, build an agent that does. The official MCP SDKs handle the protocol; CMP42 provides the tools. Your agent discovers what is available at runtime — including tools for models you added last week — and calls them with typed inputs.
Writes are validated against your models and attributed to your agent, so a custom agent leaves the same audit trail as any other client. Put it inside a Process and you also get approval hooks, run history, replay and token cost per run.
The snippet uses the TypeScript MCP SDK to illustrate the pattern; connection details for CMP42 ship with the Alpha.
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
const client = new Client({ name: "renewal-agent", version: "0.1.0" });
await client.connect(transport); // transport to your CMP42 workspace
const { tools } = await client.listTools();
// → get_context, list_related, write_note, get_contact, …
const deals = await client.callTool({
name: "list_deals_at_stage",
arguments: { stage: "renewal" },
});NOT MCP? REST + GRAPHQL.
For non-MCP clients, every operation is exposed through REST and GraphQL — same auth, same audit trail. Ship the agent of the month against the same surface.
Your team’s Slack bot
A bot your team runs can answer “what is the status of Aspen?” from the graph and write a note back — through REST or GraphQL, attributed to the bot.
The app you build next month
Internal tools, client portals, the dashboard you actually want. Query exactly the slice of the graph you need with GraphQL, or keep it simple with REST.
The agent of the month
A new agent framework ships every week. Whatever it speaks — MCP or plain HTTP — it meets the same models, the same validation and the same audit trail.
Two native, CMP42-maintained integrations are planned — neither ships in the Alpha, both are scheduled for the Beta in Q1 2027. A Slack app (workspace-level bot plus slash commands) for teams that already run their operations in Slack, and a ChatGPT connector (custom GPT with an MCP bridge) for teams that use ChatGPT as their agent surface. Beyond those two, every other integration is expected to run over the generic MCP, REST or GraphQL layer rather than as a vendor-specific coupling — you get the same auth model and the same audit trail whichever surface you use.
BRING THE MODEL YOU TRUST.
We do not ship a proprietary AI feature layer. CMP42 gives your agent grounded context; your agent produces the insight. If we shipped a black box, we would be part of the problem we set out to fix.
In practice that means:
- You choose the model — per team, per task.
- You can switch without migrating your CRM. The graph, the models and the audit trail stay where they are.
- You control the data flow. An agent sees only what it reads through the tools or the API, and on the managed EU tier nothing leaves except what you explicitly send to an agent you connect.
SHAPE THE TOOL SURFACE.
Design Partners can propose and name one MCP tool. If it fits the platform, we build it and ship it to every future customer — designed around a real workflow with the agents you actually use. Twelve slots, six months of Alpha from October 2026, €6,900 one-time.