Skip to main content
Add a server to config.toml and its tools join your catalog as <name>__<tool>, ready for agents and plan steps.

​
Before you start

You need the server’s launch command or its URL. If it needs a token, put the token in an environment variable rather than in the file.

​
Add the server

1

Add an [mcp.<name>] table

The table name becomes the prefix of every tool from the server. Use command for a server graph starts itself over stdio, or url for a streamable HTTP server.
description is optional, but setting it generally improves how well the planner chooses the server’s tools. Values in env and headers expand ${VAR} from your environment. If a variable isn’t set, graph doesn’t connect to that server and names the missing variable.
2

Test the connection

​
Check it worked

List the server’s tools, then call one through graph:
graph returns a tool’s structured content when the server sends it. Otherwise it parses the text the tool returns as JSON, or wraps plain text as {"text": ...}, as here. To load only some of a server’s tools, or fix a tool’s description, see Filter and override a server’s tools. If the server won’t connect, see Test an MCP server.
​
object
MCP servers, as [mcp.<name>]. Their tools join the catalog as <name>__<tool>.
​
string
What the server is for, read by drafters choosing tools.
​
string
Executable for a stdio server. Use command or url, not both.
​
string[]
Arguments for command.
​
map<string, string>
Environment for a stdio server. Values expand ${VAR}.
​
string
Endpoint for a streamable HTTP server.
​
map<string, string>
HTTP headers. Values expand ${VAR}.
​
string[]
Load only these tools.
​
string[]
Skip these tools.
​
object
Per-tool corrections, as [mcp.<name>.tool_overrides.<tool>].
​
string
Replaces the server’s description of the tool.
​
object
The tool’s output shape.
​
object
An example of the tool’s output.
Go to full reference