Unfour
Sign in 中文
All features

MCP Integration

Let Codex and Cursor use your backend workspace through MCP

Unfour MCP is a local stdio MCP server for Codex, Cursor, and compatible clients. It exposes approved API, SSH, and Database tools using the same saved workspace resources as the desktop UI; Flow tools manage and execute the same local runbooks.

For developers using Codex, Cursor, or another MCP-compatible coding agent who need to check whether a change worked in the target environment, under workspace policy and user control.

Read the docs

Coding tools + runtime tools

Coding agent · its own tools

  • Inspect the repository and read code
  • Edit repository code
  • Run code and repository tests

Unfour MCP · runtime tools

  • Send or replay API requests
  • Inspect SSH/server evidence
  • Query database state
  • Inspect workspace and environment context
  • Manage and run local Flow runbooks

The coding agent edits the repository with its own coding tools. Unfour supplies the controlled runtime side of the investigation; it is not an IDE, does not manage repository code, and does not run a full coding agent.

What MCP adds to the Unfour workspace

The local stdio MCP server lets an agent move from code-level reasoning to API, server, and database evidence through Unfour’s command and policy layers.

API evidence

Inspect saved requests and redacted history, review environments, and send policy-allowed requests.

SSH and runtime evidence

Review command history, run scoped diagnostics, inspect remote files, and work with saved SSH tasks.

Database evidence

List saved connections and schemas, run read-only queries or explain output, and use policy-controlled execution.

Shared Flow runbooks

List, read, save, delete, run, or cancel local Flows and inspect run history. The tools share the desktop definitions and resources; running a Flow requires confirmation even in full access and is blocked by read-only policy.

Workspace and environment context

Resolve the active workspace, environment, variables, connections, and recent local activity before acting.

Policy and confirmation

Tool calls are checked against workspace policy. When policy or backend safety requires confirmation, approval is bound to the exact action content.

MCP client setup

Use guided setup for Codex and Cursor, or manually configure another MCP-compatible client that supports a local stdio server.

Example workflow

Change, investigate, then verify runtime behavior

Codex or Cursor uses its own tools for reviewed code changes. With the changed service running in the intended environment, use Unfour MCP to repeat requests, inspect SSH logs, and query database state. Flow tools can execute saved local runbooks for known checks; interpretation and remediation remain user-directed.

  1. Code change · agent’s own tools
  2. Reproduce API behavior · Unfour MCP
  3. Inspect SSH / runtime evidence · Unfour MCP
  4. Check database state · Unfour MCP
  5. Agent proposes the next change · developer reviews
  6. Re-run the original check · Unfour MCP
  7. Verify the response and relevant database state

AI-assisted, not fully autonomous

The local workspace defines what the agent can do

MCP operations run through the local Unfour workspace. Tools are explicit, scoped, redacted where needed, and subject to policy before the agent receives evidence or performs an action.

MCP blocks secret-reveal operations; scoped masking does not guarantee that arbitrary tool output contains no confidential text.

Tools operate on saved, workspace-scoped resources; Flow definitions and history remain local-only.

Auto policy maps dev to full access, test to guarded, and prod to read-only. Explicit policy overrides auto; full access does not guarantee a prompt for every risky action.

The AI agent gathers evidence and proposes next steps; the user remains in control.

Unfour MCP questions

Short answers about compatibility, credentials, permissions, and control.

What is Unfour MCP?

Unfour MCP is a local stdio server exposing approved API, SSH, Database, workspace, activity, and system tools. Codex and Cursor use the same saved resources as the desktop UI. Flow tools manage and execute local runbooks.

Which AI agents can use Unfour MCP?

Unfour provides guided configuration for Codex and Cursor. Other MCP-compatible clients can connect when they support a manually configured local stdio server.

Do SSH or database credentials leave my machine?

SSH and database credentials use local secret storage and are resolved on the device for requested connections. MCP blocks secret-reveal operations, but logs, query results, and other free text can still contain confidential information. Review what you expose to the connected AI client.

Can an AI agent execute any command automatically?

No. Through Unfour MCP, agents can call only exposed tools within workspace policy. Policy or backend safety can require content-bound confirmation; full access does not guarantee a prompt for every risky action.

Can Codex or Cursor manage and execute Unfour Flow?

The Flow tools let Codex and Cursor manage local definitions, start and cancel runs, and inspect shared run history. Flow execution requires confirmation even with full access and is blocked by read-only policy. Flow V1 is local-only and is not cloud synced.

Does Unfour automatically fix production issues?

No. Unfour MCP supports a diagnose-then-act workflow, but it does not run an autonomous troubleshooting playbook. The developer reviews evidence and decides the fix.

Related features

Connect your AI agent to backend evidence, not unrestricted access

Download Unfour and configure the local MCP server from the desktop application.

Download Unfour Read the docs View all features