Filter and override a server's tools

Choose which of an MCP server’s tools graph exposes, and correct their descriptions and output shapes.

These settings go in the server’s [mcp.<name>] table in config.toml. They change what graph shows agents and drafters, not what the server does.

config.toml

[mcp.everything]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-everything"]
include_tools = ["echo", "get-sum"]

[mcp.everything.tool_overrides.echo]
description = "Returns the message it is given, prefixed with 'Echo: '."
output_schema = { type = "object", properties = { text = { type = "string" } } }
output_example = { text = "Echo: hello" }

Limit which tools load

List the tools you want in include_tools, or the ones you don’t in exclude_tools. Use the server’s own tool names, without the <name>__ prefix. When you set both, graph loads the included tools that aren’t excluded.

[mcp.everything]
include_tools = ["echo", "get-sum"]

Every tool you load is offered to agents and drafters, so leaving out the ones you don’t use keeps their prompts smaller and their choices easier.

graph mcp tools everything

everything — 2 tools
  echo [read-only]
  Returns the message it is given, prefixed with 'Echo: '.
  get-sum [read-only]
  Returns the sum of two numbers

Override a tool’s description

Agents and the planner choose tools by their descriptions. When a server’s description is vague, or two tools are hard to tell apart, replace it in a tool_overrides table named after the tool:

[mcp.everything.tool_overrides.echo]
description = "Returns the message it is given, prefixed with 'Echo: '."

Declare a tool’s output shape

Many servers don’t say what their tools return. Add output_schema, output_example or both to the same override table so drafters can wire the tool’s result into later plan steps, and plan validation can check references to it. A server’s own output schema is replaced by yours.

[mcp.everything.tool_overrides.echo]
output_schema = { type = "object", properties = { text = { type = "string" } } }
output_example = { text = "Echo: hello" }

graph tool show everything__echo prints the description and output schema now in effect. graph also learns a tool’s output shape as it runs; see Improve what graph knows about a tool.