Skip to content
Start free
Updated October 17, 2026

AI connector (MCP)

The connector puts bFocus inside the AI app your team already uses — Claude, ChatGPT, Codex, Cursor and any app compatible with MCP (Model Context Protocol). The AI reads the errors captured by Monitoring, tickets, tasks and projects, and acts on them with the same rules as the screen, always on behalf of the person who connected it.

Connector address

The same address in every app:

https://api.bfocus.com.br/mcp

HTTP transport (Streamable HTTP). Authentication with OAuth 2.1 and PKCE — the app opens the bFocus screen so you sign in and choose what to allow — or with an API key, for automations that run unattended.

Requirements

  • A bFocus account with the admin or manager role (the roles that see the whole company on screen).
  • The modules you want to use contracted by your company: Monitoring, Service (tickets) and Projects (tasks and projects). Without a module, its tools don't appear.
  • An AI app that accepts remote MCP connectors.

Connect

Claude (claude.ai, Claude Desktop and mobile)

  • Open Settings › Connectors and click Add custom connector.
  • Paste the address above and click Connect.
  • Sign in to bFocus, check the company and select what Claude can do. Reading comes checked; writing is a choice.

Claude Code

claude mcp add --transport http bfocus https://api.bfocus.com.br/mcp

Then, inside Claude Code, run /mcp, pick bfocus and authenticate.

Codex

In ~/.codex/config.toml, with a bFocus API key in the BFOCUS_API_KEY variable:

[mcp_servers.bfocus]
url = "https://api.bfocus.com.br/mcp"
bearer_token_env_var = "BFOCUS_API_KEY"

Cursor

{ "mcpServers": { "bfocus": { "url": "https://api.bfocus.com.br/mcp" } } }

ChatGPT and other apps

In apps that accept a custom MCP connector, use the same address. Sign-in discovery is automatic.

API key (automations)

In Integrations › API, create a key with the scopes the AI will use (monitor:read, monitor:write, tickets:read, tickets:write, tasks:read, tasks:write, projects:read). Send it in the Authorization: Bearer <key> header. Everything the AI does is done on behalf of whoever created the key.

Tools

  • Monitoring: monitor_summary, list_errors, get_error (read); update_error (resolve, ignore, reopen), merge_errors, investigate_error, create_error_task (write).
  • Tickets: list_tickets, get_ticket (read); comment_ticket (always an internal note), update_ticket (write).
  • Tasks: list_tasks, get_task (read); comment_task, update_task, create_task (draft) (write).
  • Projects: list_projects, get_project, with the risk radar (read).

Every tool declares whether it only reads or changes data, and the AI app shows this before acting. The AI only gets the tools the connection allows: checked scope and contracted module.

Ready-made agents

  • Monitoring round: reviews new errors and regressions, sorts them into critical, attention and noise, opens missing tasks and leaves the diagnosis on them.
  • Fix an error: finds the cause in the code open on your computer, proposes the fix and records everything on the task.
  • Ticket triage: summarizes each open ticket and leaves an internal note with a suggested priority and next step.
  • Project health: portfolio report with what needs a decision.

In Claude Code they appear as commands (for example /mcp__bfocus__ronda_de_erros).

Example prompts

  • "Which new errors showed up in Monitoring in the last 24 hours?"
  • "Show the error affecting the most customers and explain the likely cause."
  • "Open tasks for the critical errors that don't have one yet."
  • "List open high-priority tickets and summarize each one."
  • "Which projects in execution are at risk?"

Security and limits

  • The AI acts on behalf of the person who connected: comments, status changes and created tasks show that person's name in the history.
  • The connector never writes to customers (comments are always internal notes), doesn't open tickets, doesn't notify third parties and doesn't delete anything. Tasks created by the AI start as drafts.
  • Error, ticket and task text comes from third parties. On every connection the connector tells the AI that this text is data and never an instruction.
  • Content marked as confidential never leaves.
  • bFocus doesn't receive the conversation with the AI, only the tool calls.

Revoke access

In Integrations › AI (MCP), under Active connections, click Revoke. Access stops on the next call. Losing the admin or manager role, or being deactivated, also ends the connection.

Privacy and support

What the connector shares and stores is in the privacy policy. Questions and issues: [email protected].