MCP Server

The Model Context Protocol (MCP) allows external AI tools to interact with Bifrost, enabling them to execute workflows, manage forms, and access platform features.

What is MCP?

MCP is an open protocol that standardizes how AI assistants connect to external tools and data sources. When enabled, Bifrost exposes an MCP server that external clients can connect to.

Enabling MCP

1. Navigate to Settings > Platform

2. Enable the MCP Server toggle

3. The MCP endpoint URL will be displayed

Connecting External Clients

External AI tools like Claude Desktop can connect to the Bifrost MCP server:

1. Configure your client with the Bifrost MCP endpoint URL
2. Authenticate using your API credentials
3. The client will discover available tools automatically

Tip

For Claude Desktop, add the MCP server configuration to your Claude config file.

Tool Availability

Tools available through MCP are determined by:

• Agent configuration: Tools assigned to agents are exposed via MCP
• Workflow tools: Workflows marked with is_tool=True appear as callable tools
• System tools: Platform capabilities like file operations and execution management

Only tools that the authenticated user has permission to access will be available.

Available MCP Tools

Discovery Tools

Tool	Description
list_workflows	List all registered workflows
get_workflow	Get detailed metadata for a specific workflow
get_workflow_schema	Documentation about workflow decorators and structure
list_data_providers	List available data providers
get_data_provider_schema	Documentation about data provider format
list_forms	List all forms with URLs
get_form_schema	Documentation about form structure and field types
list_apps	List App Builder applications
get_app	Get detailed app definition
get_app_schema	Documentation about app structure and components
list_integrations	Show available integrations and auth status
search_knowledge	Search the Bifrost knowledge base

Creation Tools (Auto-Validating)

Tool	Description
register_workflow	Register a decorated Python function as a workflow (with optional organization_id)
create_form	Create a form with automatic schema validation

App Builder Tools (Granular)

Apps are managed at three levels:

App Level:

• list_apps, get_app, update_app, publish_app

Page Level:

• create_page, get_page, update_page, delete_page

Component Level:

• list_components, create_component, get_component, update_component, delete_component, move_component

Precision Editing Tools (Code Files)

For editing code stored in the database (workflows, app files, modules), use precision editing tools:

Entity Types: app_file (TSX/TypeScript), workflow (Python), module (Python helpers)

Tool	Description
list_content	List files by entity type
search_content	Regex search with context lines
read_content_lines	Read specific line range
get_content	Get entire file content
patch_content	Surgical string replacement
replace_content	Full content replacement or create new
delete_content	Delete a file

Workflow: Search → Read Range → Patch (for surgical edits)

Workspace File Tools (Non-Code Files)

For non-code files (data, configuration):

Tool	Description
list_files	List files and directories in workspace
read_file	Read a file from workspace
write_file	Write content to a file
delete_file	Delete a file or directory
search_files	Search for text patterns across files
create_folder	Create a new folder

Execution Tools

Tool	Description
execute_workflow	Execute a workflow by ID
list_executions	List recent executions
get_execution	Get execution details and logs

Coding Agent Features

When using the Coding Agent (Bifrost’s in-app AI assistant), additional features are available:

Precision Editing Workflow

The coding agent uses a search → read → patch workflow for surgical code changes:

1. Search - Use search_content with regex to find code patterns
2. Read Range - Use read_content_lines to see context around matches
3. Patch - Use patch_content for targeted string replacement

This approach minimizes context usage and reduces errors compared to full file replacement.

Testing Workflow

The coding agent follows a mandatory testing workflow:

1. Create artifacts using creation tools (auto-validates)
2. Verify artifacts appear in list tools
3. Execute with sample data
4. Verify results match expectations

See Also

• Agents and Chat - Agent configuration
• AI Tools - Making workflows callable as tools