Enrich and store contacts in a personal CRM
Using Every's agent-native architecture to create my dream digital rolodex. Task: The author built a personal CRM using an agent-native architecture where Claude acts as the interface, connected to a Supabase database via MCP with four atomic tools (query, insert, update, delete). Markdown files define workflows, allowing the agent to compose tool calls based on natural language prompts. The system integrates Harmonic for profile enrichment and Granola for meeting note attachment, automatically flagging new contacts from transcripts. The human role is limited to initiating commands and reviewing suggestions, as the agent handles data retrieval and storage logic.; tools: Claude. Human role: The original post does not say how a person stayed involved.; limitations: The implementation relies on specific third-party services (Harmonic, Granola) and an MCP-compatible database (Supabase), introducing integration complexity and potential cost. The source does not detail error handling for failed enrichments or data conflicts. Human approval is implied for adding contacts but not explicitly defined as a mandatory gate. The reliability of the agent's autonomous tool composition is not tested or quantified in the text..
Tools
Published on Agent Work Atlas:
Overview
Using Every's agent-native architecture to create my dream digital rolodex
Workflow
The author built a personal CRM using an agent-native architecture where Claude acts as the interface, connected to a Supabase database via MCP with four atomic tools (query, insert, update, delete). Markdown files define workflows, allowing the agent to compose tool calls based on natural language prompts. The system integrates Harmonic for profile enrichment and Granola for meeting note attachment, automatically flagging new contacts from transcripts. The human role is limited to initiating commands and reviewing suggestions, as the agent handles data retrieval and storage logic.
Reported result
The source describes the system's capabilities and workflow but does not provide measured outcomes, performance metrics, or evidence of successful execution beyond the author's narrative. It claims profiles appear in seconds and search returns ranked lists, but these are qualitative descriptions without data. No failures or specific business impacts are reported. A measurable acceptance check would require verifying the latency of profile creation and the accuracy of search results against a known dataset.
Human role
The original post does not say how a person stayed involved.
Limitations
The implementation relies on specific third-party services (Harmonic, Granola) and an MCP-compatible database (Supabase), introducing integration complexity and potential cost. The source does not detail error handling for failed enrichments or data conflicts. Human approval is implied for adding contacts but not explicitly defined as a mandatory gate. The reliability of the agent's autonomous tool composition is not tested or quantified in the text.
Try it
Act as a CRM automation agent. I will provide a contact name and a LinkedIn URL. First, use the Harmonic tool to fetch their role, company, and skills. Second, check the Granola tool for any meeting notes mentioning this person. Third, query the Supabase database to see if the contact already exists. If not, insert the enriched profile and attach any relevant notes. If they exist, update the record with new information. Finally, suggest any other contacts found in the notes who are not yet in the CRM. Before executing any insert or update, display the proposed changes and wait for my confirmation. Start with a dry run on [replaceable input: test contact name and URL] to verify the workflow.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.