Advanced MCP Configuration¶
Ralph Workflow is a free and open-source AI agent orchestration system built around a simple Ralph-loop core. That simple core composes into a stronger composable workflow system for substantial, well-specified repo work, and the default workflow is already strong enough to start with before you customize anything.
This page covers Ralph Workflow’s tool and integration layer: web search, readable-page fetches, browser automation, and custom MCP servers. Use it when you want to extend the default workflow’s reach without muddying the core workflow docs or leaking provider/tool setup into the wrong page.
This page is for operators who want to control Ralph Workflow’s tool and external-integration layer. Use it when you need to wire in search, readable-page fetches, or custom MCP servers without muddling that work into the main workflow docs.
This is also where the trust boundary matters most. Prefer plugging Ralph Workflow into tools you already trust and keep secrets out of repo-local config whenever you can.
Which file am I editing?¶
project-local MCP overrides →
.agent/mcp.tomluser-global MCP defaults →
~/.config/ralph-workflow-mcp.tomlbundled default / example →
ralph/policy/defaults/mcp.toml
Use project-local config when a repo needs custom MCP servers. Use user-global config when you want the same servers available everywhere.
What mcp.toml controls¶
mcp.toml configures:
MCP servers over
stdioorhttpsearch backends
web-visit / readable-page fetch behavior
media handling toggles
advanced crawling integrations
Major sections¶
[mcp_servers.<name>]¶
Defines a named MCP server.
Common fields:
transportcommandargsurlenv
[web_search]¶
Controls whether search is enabled and which backend/fallback chain is used.
[web_search.backends.<name>]¶
Backend-specific configuration.
[web_visit]¶
Controls readable-page fetch behavior.
Typical fields include:
enabledtimeout_msmax_bytesuser_agentallow_private_networksextract_links
[media]¶
Controls broad multimodal file reading behavior.
Common advanced user stories¶
I want to add a GitHub/docs/custom MCP server¶
Add a new [mcp_servers.<name>] block.
I want different MCP servers in one repo than in the rest of my machine¶
Use .agent/mcp.toml.
I want web search enabled with explicit backends¶
Edit [web_search] and the backend-specific blocks.
I want to wire in Crawl4AI or another HTTP MCP service¶
Add an http MCP server block pointing at the service URL.
Safety rules¶
keep this file secret-free where possible
prefer
api_key_envor environment variables over inline secretsdo not commit real credentials into repo-local
.agent/mcp.toml