Skip to main content

Customizing tools

As we saw in the how it works section, Exograph's MCP server offers the execute_query tool in combined mode and an additional introspect tool in separate mode. This is often sufficient when using Exograph directly from Claude Desktop or similar MCP clients. However, when using Exograph as part of an agentic workflow, you may want finer control over the tools to focus on specific parts of the schema.

Profiles allow you to specify which types, queries, and mutations are exposed through MCP tools.

Specifying profiles​

By default, all queries and no mutations are exposed. You can customize this behavior by creating profiles. In agentic workflows, this allows the agent to examine the prompt and context to select relevant tools. For example, if the prompt is "Get me the list of branches", there's little point in offering customer-related queries to the LLM. When you anticipate specific prompt patterns (often governed by evaluation criteria), you can create profiles that improve agent performance.

tip

Profiles relate closely to bounded context from domain-driven design. You form boundaries around related domain parts and expose only tools specific to that bounded context. Agents can then choose appropriate tools based on the prompt and context.

You specify profiles in the exo.toml file at the root of your project (for other usages of the exo.toml file, see here).

[[mcp.profiles]]
name = "branch_management"

For each profile, Exograph exposes the execute_query_<profile_name> tool (and corresponding introspect tool in separate mode). The above profile exposes the execute_query_branch_management tool (and introspect_branch_management tool in separate mode). If you don't specify other attributes, the tool exposes all queries and no mutations.

You can control exposed queries and mutations using include/exclude patterns. Each property accepts an array of wildcard patterns. For example, "branch*" matches branches, branch, branchAgg, etc. Use "*" to match all types or operations.

Query Control Properties:

  • queries.operations.include: Include queries matching pattern (default: ["*"] - all queries)
  • queries.operations.exclude: Exclude queries matching pattern (default: [] - none excluded)
  • queries.models.include: Include models matching pattern (default: ["*"] - all models)
  • queries.models.exclude: Exclude models matching pattern (default: [] - none excluded)

Mutation Control Properties:

  • mutations.operations.include: Include mutations matching pattern (default: [] - none included)
  • mutations.operations.exclude: Exclude mutations matching pattern (default: [] - none excluded)
  • mutations.models.include: Include models matching pattern (default: [] - none included)
  • mutations.models.exclude: Exclude models matching pattern (default: [] - none excluded)

Let's see how to use each of these properties.

Restricting queries by name​

You can restrict queries by name using the queries.operations.include and queries.operations.exclude properties. For example, to expose only the branches query and all queries starting with customer:

[[mcp.profiles]]
name = "branch_management"
queries.operations.include = ["branches", "customer*"]

This configuration exposes the branches query and all queries starting with customer while excluding others like branch, branchAgg, and financialAdvisors.

Alternatively, you can exclude queries by name. For example, to expose all queries except aggregate queries:

[[mcp.profiles]]
name = "non_aggregate_queries"
queries.operations.exclude = ["*Agg"]

You can combine both properties to create a profile that exposes only the branches query and all queries starting with customer, except for aggregate queries:

[[mcp.profiles]]
name = "branch_management_non_aggregates"
queries.operations.include = ["branches", "customer*"]
queries.operations.exclude = ["*Agg"]

Query names refer to GraphQL operation names. For Postgres modules, query names follow the naming conventions described in queries and mutations.

Restricting queries by type​

Often, you'll want to control queries based on their domain model using the queries.models.include and queries.models.exclude properties.

For example, to expose only queries related to the Branch type:

[[mcp.profiles]]
name = "branch_management"
queries.models.include = ["Branch"]

This configuration exposes all queries that return a single Branch or list of Branch entities, such as branch, branches, branchAgg, etc.

You can also exclude models by name. For example, to exclude all models starting with Membership:

[[mcp.profiles]]
name = "non_membership_queries"
queries.models.exclude = ["Membership*"]

You can combine all these properties to create profiles that control exposure based on both names and types.

Restricting mutations​

Controlling exposed mutations works similarly. However, by default, Exograph doesn't expose any mutations. You can enable mutations selectively by creating profiles.

For example, to enable specific customer management operations while excluding delete operations:

[[mcp.profiles]]
name = "customer_management"
mutations.models.include = ["Customer"]
mutations.operations.include = ["createCustomer", "updateCustomer"]
mutations.operations.exclude = ["deleteCustomer"]

This profile exposes all Customer-related mutations while specifically excluding the deleteCustomer operation, providing a safe subset for customer management workflows.