Agentic platform

Your revenue data,
in the agent you already use.

Dopia is an MCP server. Point Claude Code, Claude Desktop, Codex or anything else that speaks MCP at one URL, approve the scopes, and the agent can read your workspace, write records back, and publish what it finds as a page you can send.

Start freeRead the setup guide
https://mcp.dopia.ai/mcp

Three consents, and you pick which.

Scopes are not a checklist for a security page. They decide which tools the agent is even told about.

crm:read

Read the workspace

Look up a company, pull a record, count what is in the pipeline — with the same table schema the product uses, not a flattened export.

  • search_crm
  • search_records
  • get_record
  • get_table_schema
  • aggregate_records
crm:write

Write records back

Create and update records. Identity resolution is the same code path the product runs, so an agent cannot quietly mint a duplicate company.

  • create_record
  • update_record
artifacts:write

Publish an artifact

Turn a finished analysis into a page with its own URL that you can send to someone who has no account. Held apart from crm:write on purpose — publishing and record-writing are different consents.

  • create_artifact
Clients

One URL. Whatever you already have open.

There is nothing to install on our side and no per-client integration to wait for. If it speaks MCP over streamable HTTP, it connects — and the same workspace answers all of them, so research started in a terminal is on the record by the time you open the app.

  • Claude Code

    Terminal, IDE and desktop app

  • Claude Desktop

    Connectors

  • Codex

    MCP client configuration

  • Any MCP client

    Streamable HTTP transport

An agent with your CRM is a security question.

So here is the answer, before you ask it.

OAuth, not a pasted key

You approve the connection in the browser and see, in plain words, what it is being allowed to do before you approve it.

Two gates, not one

A credential’s scopes decide which tools are advertised, and every call is re-checked at execution. An out-of-scope call never runs, even if the model invents it.

One kitchen, two counters

MCP tools are the same implementations the in-product agent uses. There is no second, thinner copy of your CRM logic to drift out of sync.

Pick the one you already have open.

Same URL, same consent screen, same tools. The only thing that differs is the two lines you type to get there.

Run it with Claude

Claude Code in the terminal or your IDE, and Claude Desktop as a custom connector.

Run it with Codex

One command, or three lines in ~/.codex/config.toml if you keep configuration in a file.

Connect it to the agent you already trust.

One URL, scopes you approve in the browser, and a workspace that keeps what your agents find instead of losing it with the conversation.

Start freeTalk to sales

MCP over streamable HTTP · OAuth · revoke any time