CorpusIQ is now live in the ChatGPT app store.
Find CorpusIQ in ChatGPT
Connector directory
Files & Storage
Advertising & Analytics
CRM
Email Marketing
Commerce
Finance
CorpusIQ is now live in the ChatGPT app store.
Find CorpusIQ in ChatGPTConnector directory
For Claude users with business systems
One governed MCP connection gives Claude scoped access to every business system you approve. No local servers, no config files, no credential spread across machines.
Governed OAuth · 30-day free trial · Cancel anytime
Each of these reaches live systems and returns the records behind the answer.
The first server is easy. The tenth is the problem.
Wiring Claude to your business one server at a time means a separate process, a separate credential, and a separate failure mode for every system. Five tools is a config file. Twenty is an operations problem nobody owns.
Each locally configured server holds its own token. Rotating them, revoking one cleanly, or answering which machine currently holds access to the books becomes guesswork.
A server running on a laptop serves that laptop. Colleagues on Claude web, another device, or a different assistant see none of it, so the setup has to be repeated per person and per machine.
When the CRM and the accounting system disagree about a number, a pile of independent servers has no view about which one is authoritative. Whichever answers first wins, which is not the same as being right.
What changes when the connection is consolidated and remote.
A single MCP endpoint fronts every system you authorize, so Claude gets one connector to manage instead of one per tool. 40+ connectors and 650+ tools are reachable through it.
Each external source is authorized separately through its own login screen. Disconnecting one from CorpusIQ removes its stored connection state and requires reauthorization before reuse; the other connections remain untouched.
Claude sees named tools with typed arguments, not a pile of undifferentiated API surface. That is what makes it pick the right call for a vague question instead of guessing at parameters.
Responses carry the records behind them. When two systems could answer the same question, the answer says which one it used, so a disagreement is visible rather than silently resolved.
Six steps, all inside Claude. Nothing is installed on your machine.
Go to the connector settings in Claude. The connection is configured in Claude itself, so nothing is installed on your machine and nothing runs locally.
Choose to add a custom connector. This is Claude's built-in path for remote MCP servers, which is what CorpusIQ is.
Use https://mcp2.corpusiq.io/mcp. The transport is streamable HTTP, so any MCP client that supports remote servers can use the same endpoint.
Sign in to CorpusIQ when prompted. Claude holds the resulting credential, so there is no key to paste into a config file or keep in sync by hand.
Approve each external source you want reachable through its official login screen. Source access is scoped per system. Disconnecting a source from CorpusIQ removes its stored connection state and requires reauthorization before reuse.
The tools appear in Claude immediately. Ask in plain English and the answer comes back with the records it drew on.
Add a tool by authorizing it, not by editing a config and restarting a process. The MCP connection in Claude does not change when your connected systems do.
The server is remote, so the same connection is there on web, desktop, and other devices. There is nothing installed to fall out of date on one machine.
The endpoint is a standard remote MCP server, so the same authorized systems are reachable from ChatGPT, Perplexity, Slack, and any MCP client you point at it. Authorize once, not once per assistant.
Read-only external-source retrieval tools do not write back to vendor systems. The only tools that accept writes are CorpusIQ control-plane tools, which are separately named, carry explicit safety annotations, and act on your CorpusIQ configuration rather than a connected system; control-plane actions include updating or removing user-declared facts, decisions, metric specifications, and source manifests.
Disconnect a single source without disturbing the others, or disconnect the whole CorpusIQ connector at once. Disconnect removes CorpusIQ's stored connection state; provider-side authorization remains governed by that provider's own controls.
Each tool call reads the connected source at request time, so answers reflect current provider records rather than a scheduled warehouse sync.
One connection does not mean broad access. Every external source is authorized separately and can be disconnected from CorpusIQ on its own.
External source connectors authenticate through the provider's own OAuth flow and use the documented retrieval scopes. CorpusIQ never asks for the connected service's password.
For direct MCP connector requests, CorpusIQ fetches source records live and returns them to the requesting AI client. The direct path does not retain raw customer files or full connector response payloads, and it does not build embeddings, file indexes, or cached or indexed summaries. Local AUDIT logs record raw query text and tool parameters plus bounded result summaries. In Azure, the Log Analytics workspace retains those logs for 30 days. The selected AI client's plan and settings govern conversation handling after it receives the data. Optional indexed search is separate and retains embeddings and minimal metadata until connector revocation or account deletion.
Your AI provider's plan policy governs its conversation handling. SOC 2 aligned, CASA Tier 2 certified by DEKRA, with the certificate published.
Each connector lists the exact scopes it requests before you approve it. Read the full security posture.
https://mcp2.corpusiq.io/mcp. It is a remote MCP server over streamable HTTP, so it works with Claude's custom connector flow and with any other MCP client that supports remote servers.
No. The server is remote, so there is no local process, no config file, and nothing to keep running. That is also why the same connection works across your devices instead of only the machine you set it up on.
Read-only external-source retrieval tools do not write back to connected vendor systems. The only tools that accept writes are CorpusIQ control-plane tools, which are separately named, carry explicit safety annotations, and act on your CorpusIQ configuration rather than a connected system; control-plane actions can update or remove user-declared facts, decisions, metric specifications, and source manifests when you request them.
One connection instead of many, one credential path instead of one per server, and stored connection state you can disconnect per system from a single place. It also arbitrates when two connected systems could answer the same question, which independent servers cannot do.
Yes. The systems you authorize are reachable from ChatGPT and Perplexity, inside Slack, and from any MCP client pointed at the same endpoint. Microsoft Teams is coming soon. You authorize a tool once rather than once per assistant.
Adding the connector in Claude takes about a minute. Each business system is another two minutes or so through its own login screen. 30-day free trial, no credit card.
The protocol behind the connection, explained without the marketing.
Endpoint, transport, headers, and the device-login flow for other MCP clients.
What CorpusIQ reads on Claude's behalf, connector by connector.
The same connected systems, reached from ChatGPT instead of Claude.
What the connection looks like against a single system, the accounting file.
The cross-system reporting view, for questions that span several tools at once.
30-day free trial. Governed source access. Cancel anytime.