Planhat

Breadth across classes, with a native API and an MCP server behind it rather than a fixed connector list. Named integrations span CRM in Salesforce and HubSpot, billing in Stripe, data warehousing in Snowflake, support in Intercom and Jira, and collaboration in Slack.

Underneath sits a REST API at api.planhat.com with bulk upsert for custom data and a separate high-throughput analytics and tracking endpoint for usage telemetry, so a customer models their own objects into the platform rather than accepting a fixed schema, which is the point of a system that deploys as CSP, CRM or PSA over one data model. Outward, the MCP server exposes the same live customer data to connected assistants under customer-set permissions.

Sourceplanhat.com/applications/crm and planhat.com/developers/ai/mcp-serverread 2026-09-04

Lifecycle automation and playbooks running across a single operational workspace that can be a customer success platform, a CRM and a professional services tool at once, so a chain crosses functions without crossing systems.

The vendor's own framing is that the platform is architected to equip agents and humans alike to execute with clarity and control rather than only to report, and agents self-configure around the customer's documented business rules, so what a play does is shaped by the SOPs in the model rather than by a generic template.

The chain is addressable from outside as well as inside: the MCP server exposes model routes and operations so a connected assistant can run multi-step work against live data, with tools and parameters discovered at runtime. The catch buyers raise is that the flexibility that makes the chain span three product shapes rests on a data model complex enough to need an operations owner to run it.

Sourceplanhat.com and planhat.com/developers/ai/mcp-serverread 2026-09-04

A maintained model of the customer's business the agents ground on. The living model spans CRM records, time series data, external telemetry and standard operating procedures, built for multi-product, multi-entity and usage-based revenue, and the same corpus is exposed through the MCP server under the customer's permissions. Generated output carries no documented citation.

Sourceplanhat.com and help.planhat.com AI overviewread 2026-10-01

Control over what agents may see, no documented gate on what they do. The help center states that with enterprise-grade permissioning the customer decides what data a model can access, and agents configure themselves around the customer's SOPs and business rules. No approval queue, review step, confidence threshold or hold before an agent acts is documented.

Sourcehelp.planhat.com AI overview and planhat.comread 2026-10-01

Access controls and compliance are both documented. Sign in runs through single sign-on with SAML, alongside SCIM 2.0 provisioning and role-based access control, and permissioning extends into the agent layer, where the customer decides what data a connected model may reach. The compliance set covers SOC 2 Type II and ISO 27001, GDPR and CCPA, a data processing agreement and a published sub-processor list at trust.planhat.com.

Sourcetrust.planhat.com and planhat.com/legal/gdpr-commitmentread 2026-10-01

Visibility into the customer, none into the agent. Health scoring, metrics and reporting over the data model show account state and trends. Nothing traces which play an agent ran against which account and why, what it read before acting, or what a connected assistant did through the MCP server, and no retention control over such a record is documented.

Sourceplanhat.com and help.planhat.com AI overviewread 2026-10-01

The living model is the customer record, not agent memory. CRM records, time series, telemetry and SOPs accumulate in one data model the agents read. No memory the agents keep and read back as their own state, with a scope, lifetime or delete path apart from the business data, is documented.

Sourceplanhat.com and help.planhat.com AI overviewread 2026-10-01

A named region list the buyer is offered. The GDPR commitment page states that Planhat offers hosting of customer data on infrastructure in the EU and the US, with the production environment, backups and disaster recovery hosted in the applicable region.

Sourceplanhat.com/legal/gdpr-commitmentread 2026-10-01

Shipped application shapes rather than a catalog of assets to browse. Planhat deploys as a customer success platform, a CRM or a professional services automation tool over one data model, which is a genuine choice of product configuration, and agents self-configure around the customer's own business rules and standard operating procedures rather than arriving as fixed personas.

No named prebuilt agent catalog, template gallery, play library or marketplace appears anywhere on the property, and the platform's whole proposition runs the other way, with configuration derived from the customer's own documented process rather than supplied. A system that generates its configuration from the customer's SOPs has little reason to ship a template library.

Sourceplanhat.com and planhat.com/applications/crmread 2026-09-04

Work starts from signal rather than from someone opening the product, and it can be reached from outside. Triggers fire on health score movement, on rich time series data and on external telemetry ingested into the living model, across a schema built for multi-product, multi-entity and usage-based revenue, so a usage decline or a license change in one product line is a trigger rather than something noticed later. Lifecycle automation and playbooks run against those signals continuously.

Channels reach both directions: the platform itself, connected systems including Slack, Intercom and Jira, and MCP-compatible clients such as Claude and ChatGPT, where a user can ask open questions of live portfolio data or have an assistant act under their permissions. No scheduled run cadence, inbound email watcher or webhook listener is documented by name, so automatic firing comes from the signal layer rather than a published scheduler.

Sourceplanhat.com and planhat.com/developers/ai/mcp-serverread 2026-09-04

Named makers, chosen by the vendor. The help center attributes Planhat's AI features to a connection to Vertex AI and names OpenAI alongside it, with customer data never used for training by those providers, and AI usage is metered in Planhat AI credits. No model list, routing policy, per-workspace selection or bring your own model is offered.

Sourcehelp.planhat.com AI overviewread 2026-10-01

An official remote MCP server with its own developer page, sitting on a mature API estate. Planhat operates a hosted MCP server at api.planhat.com/v1/mcp, documented on its own developer site, authenticated through Planhat OAuth clients and used to connect ChatGPT or Claude to live customer data, with the customer deciding through enterprise-grade permissioning what the model may reach.

It is discoverable and execute-capable rather than a fixed read wrapper: tools are enumerated at runtime through get_model_actions and their parameters through get_model_action_parameters, operations are addressed by model route, and the documented error surface covers token validity, permission scope, invalid operations, rate limiting and server faults. A separate demo endpoint exists for non-production testing. Beneath it sits a REST API at api.planhat.com with bulk upsert, a high-throughput analytics and tracking endpoint, and an OAuth authorization server.

Sourceplanhat.com/developers/ai/mcp-server and help.planhat.com AI overviewread 2026-09-04

Models the customer tunes against outcomes, and nothing that tests the agents. Health score design is supported during onboarding and the metrics layer is flexible enough that scoring models can be reshaped as a team learns what actually predicts churn or expansion in their book, with time series data making the effect of a change visible over a period rather than at a point. That is a mechanism aimed at the business outcome. Nothing is aimed at the agents.

No evaluation harness, scored test set, benchmark, regression check when SOPs or business rules change, or sandbox comparison of one agent configuration against another is documented, which matters more here than usual because agents self-configure from those rules, so a change to an SOP silently changes agent behavior with no way to test the effect first. A separate demo environment exists at the MCP layer for non-production testing of integrations, which is a developer facility rather than an agent evaluation surface.

Sourceplanhat.com and planhat.com/developers/ai/mcp-serverread 2026-09-04

Everything the agents touch arrives through the platform's own data model or an authenticated interface.

Work runs over the living model, reaches connected systems through documented integrations to Salesforce, HubSpot, Stripe, Snowflake, Intercom, Jira and Slack, and is addressable from outside through the REST API, the analytics endpoint and the MCP server, whose published operations are model routes against Planhat objects rather than interface actions.

No browser extension, portal navigation, form filling, headless session, screen or click automation appears on the platform pages, the developer documentation or the help center. The developer documentation enumerates exactly what the platform can be made to do from outside, and none of it is interface control.

Sourceplanhat.com/developers/ai/mcp-server and planhat.comread 2026-09-04