Skip to main content
Profound MCP gives AI assistants tools to build, manage, and run Profound Agents directly from an MCP client. Use them to create automated workflows that combine Profound data with LLM reasoning, code execution, and web search. To set up the server, refer to the Connection guides.

​
How the tools work together

Agent workflows fall into two paths: running an existing Agent, or building and publishing a new one.

​
Run an Agent

1

Find an Agent

Use list_agents to browse Agents in the organization, or list_agent_definition_templates for pre-built Agent templates.
2

Inspect inputs

Use get_agent to read the Agent’s input_schema before calling run_agent.
3

Start a run

Use run_agent with inputs that match the Agent’s input_schema.
4

Poll for results

Use get_agent_run with the agent_id and the run_id returned in the previous step, until the run reaches a terminal state.

​
Build an Agent

1

Open a build session

Use start_agent_build_session with a short, plain-language intent. It returns an agent_build_session_id. Pass that same ID, unchanged, on every build call below so the whole attempt is traced as one session.
2

Explore building blocks

Use list_agent_node_types and get_agent_node_schema to assemble a workflow graph, or list_agent_definition_templates to start from a template.
3

Create or update a draft

Use create_agent_definition or update_agent_definition with preview: true to review the plan, then set preview to false to save the draft.
4

Validate

Use validate_agent_definition to catch structural issues early, and fix them with update_agent_definition. This is a fast pre-check, not the final word: publishing runs the authoritative validation.
5

Publish

Use publish_agent_definition with preview: true to review the plan, then set preview to false to make the Agent live.

​
Use natural language when working with Claude

If you’re using Claude as the client, you can prompt it in natural language to build Agents in one go, without calling multiple tools. Here’s an example prompt and an Agent structure Claude may produce with it:
An example diagram outlining the structure of an Agent Claude generated with the prompt above

​
Behavior and safety

Agents tools can create and update Agent definitions and start Agent runs. Use preview: true (the default) on create, update, and publish tools to review changes before applying them.

​
Agents tools

These tools let you build, manage, and run Agents directly from an MCP client.
List Agents defined in the authenticated organization.
Get details of a specific Agent, including its input_schema. Use this before calling run_agent to confirm which inputs the Agent expects.
Start an Agent run. Returns a run ID you can poll with get_agent_run.Each key in inputs is a property ID from input_schema.properties. The readable label lives in that property’s title. Fetch the schema with get_agent first.Example:
Get the status and outputs of a previously started Agent run. Poll this until the run reaches a terminal state.
Open an Agent build session, the first step before building or revising an Agent. Call it before any other build tool, then pass the returned agent_build_session_id, unchanged, on every subsequent build call so the whole attempt is traced as one session. It persists nothing and does not create an Agent.
List the node types available for building an Agent graph. Returns each type’s node_type identifier, display name, and a one-line description. Call this after opening a build session, when assembling a new Agent.Inputs: none.
Get the configuration schema and examples for a specific node type. Use the returned schema to fill a node’s config correctly when building or editing an Agent graph.Example:
Browse pre-built Agent templates. Each template includes a plain-language goal, the inputs it needs, what it produces, and a skeleton workflow you can use as a starting point. Call this before building a new Agent from scratch.Inputs: none.
Read back an Agent’s full workflow graph in the same format that create_agent_definition and update_agent_definition accept. Useful for copying an existing Agent or inspecting a node type you want to replicate.
Create a new draft Agent definition. Set preview to true (the default) to see a plain-language plan and workflow diagram without persisting anything. Set preview to false to save the draft after the user confirms.
Update an existing draft Agent definition. Use this to fix validation issues after create_agent_definition or to iterate on a saved draft. Supports the same preview / apply loop as create_agent_definition.
Check whether a saved draft Agent definition is well-formed. Returns valid (boolean), issues (list of actionable errors), and, when the draft is publishable, the input_schema and output_schema the Agent will expose once live. This is a fast structural pre-check: publishing runs the authoritative validation, so a draft can pass here and still be rejected at publish. Fix any issues with update_agent_definition.
Publish a draft Agent definition so it goes live and is visible across the organization. Supports a preview step that returns a plan before applying. Call validate_agent_definition first to confirm the draft is valid.