POINT YOUR AGENT
AT REAL CONTEXT.
Every model becomes a set of MCP tools automatically. Point Claude at your workspace and it can read, write and plan against your real context in one command. No connector to build, no export to paste.
WHAT MCP IS.
The Model Context Protocol (MCP) is an open standard for connecting AI applications to the tools and data they need. Instead of every assistant inventing its own plugin format, MCP defines one way for a server to describe what it offers and for a client to use it.
An MCP server publishes a list of tools: each with a name, a description and a typed input schema. An MCP client — a desktop assistant, a coding agent, your own program — discovers those tools at runtime and lets the model decide when to call them. The model asks, the server answers with structured results, and the client can ask the human before a tool runs.
The practical effect: build one MCP server, and every MCP-compliant client can use it. The specification is public at modelcontextprotocol.io.
WHY A CRM SHOULD BE MCP-NATIVE.
Adding MCP to a CRM after the fact typically means wrapping an API that was designed around a UI. CMP42 starts from the other end: the tool surface is generated from your ontology.
Tools that speak your ontology
A generic “query records” tool leaves the model guessing what exists. CMP42 generates tools per model — get_contact, list_deals_at_stage — so the agent sees your actual business objects and their typed fields.
Typed inputs, clear errors
Tool inputs follow the model you declared, and every write is validated. When an agent sends a value that does not fit, it gets a clear error and corrects itself instead of inventing a field or a stage.
One surface for every agent
Claude today, a different client next quarter, a custom agent your team builds next month. They all consume the same tools — no per-vendor plugin to rebuild each time the agent of the month changes.
TOOLS YOUR AGENT CAN CALL.
Two kinds of tools ship out of the box. Context tools work across the whole graph. Model tools are generated for every model type — the defaults and anything you declare yourself.
| Tool | Kind | What it does |
|---|---|---|
get_context | Context | Read an object together with the context around it — the typed edges and the source objects an agent can cite. |
list_related | Context | Follow typed edges from an object: the company a contact knows, the deal they own, the doc they authored. |
write_note | Context | Write a note into the graph. Validated, versioned and attributed to the agent or human that wrote it. |
get_contact | Generated · model | Read one object of a model type — here, a contact. |
list_deals_at_stage | Generated · model | List objects of a model type filtered by a declared field — here, deals at a given stage. |
write_note_on_company | Generated · model | Write to the graph, scoped to a model type — here, a note on a company. |
Every Model generates six standard tools with a consistent naming convention: {model}.create, {model}.read, {model}.update, {model}.delete, {model}.list, {model}.search. Models with tagged fields also gain {model}.filter. Names use singular snake-case for the model and dot-suffix for the action — for example contact.create, deal.update, note.search.
The exchange on the right illustrates the pattern: a model finds the renewal deals, pulls the context around one of them, and writes back a note — three calls, every answer structured, every write attributed. Tool names follow the examples above; payload shapes are illustrative and ship with the Alpha.
→ tools/call list_deals_at_stage
{ "stage": "renewal" }
← deal cmp42://deal/… "Aspen · Q4 renewal"
owner → cmp42://contact/…
→ tools/call get_context
{ "uri": "cmp42://deal/…" }
← company, 2 contacts, 3 docs · sources cited
→ tools/call write_note_on_company
{ "company": "cmp42://company/…",
"body": "Renewal risk: budget owner changed …" }
← note written · v1 · written_by: agent:claudeEVERY MODEL BECOMES TOOLS.
You never hand-write an MCP server for your CRM. Models are declared in TypeScript-shaped YAML and migrated with a single command. When a model is migrated, CMP42 generates the matching tools, with input schemas derived from the fields you declared.
Add an Engagement model and your agent can list engagements at a stage and write notes on them — without anyone writing glue code. Migrations are backwards-compatible, so the tools agents already rely on keep working.
# models/engagement.yaml
model: Engagement
fields:
title: string
stage: "scoping" | "active" | "closed"
$ cmp42 migrate
✓ get_engagement
✓ list_engagements_at_stage
✓ write_note_on_engagementWORKS WITH ANY MCP CLIENT.
Anything MCP-compliant can use the CMP42 tool surface. That includes:
- Claude Desktop
- Claude Code
- Cursor
- Cline
- Any custom agent built with the MCP SDK
For clients that do not speak MCP — a Slack bot, an internal app, a script — every operation is also available through REST and GraphQL. All integration options →
Clients connect through one of two MCP transports. Local stdio for development and CLI setups (Cursor, Claude Code local) — the CMP42 binary is launched as a subprocess. Streamable HTTP for hosted agents (Claude Desktop, ChatGPT custom connectors, hosted Cursor). Authentication is a workspace-scoped API token placed in the MCP client config; token permissions cover one workspace and a defined set of Models. Full OAuth 2.1 lands after GA.
SAME AUTH. SAME AUDIT TRAIL.
MCP is not a side door into your CRM. The auth model and the audit trail are the same across MCP, REST and GraphQL: an agent calling write_note is subject to the same rules as a service calling the API, and its write is recorded the same way.
Every object is versioned, and every write records the human or agent that produced it. When agents run inside a Process, you also get the full run history, human approval steps and token spend per run.
NAME ONE MCP TOOL.
Design Partners get the right to propose and name one MCP tool. If it fits the platform, we build it and ship it with your naming — to every future customer. Strong applications describe a real tool from a real workflow.
Twelve slots, six months of Alpha from October 2026, €6,900 one-time.