> ## Documentation Index > Fetch the complete documentation index at: https://talktohumans.app/docs/llms.txt > Use this file to discover all available pages before exploring further. # ✨ Use with your AI tools > Find contacts, prepare outreach drafts, and build team reports with TalkToHumans MCP. TalkToHumans connects compatible AI clients such as Claude, ChatGPT, Codex, and Cursor to your LinkedIn workspace through MCP. Use your AI tools to search your network, prepare drafts for review, and turn team activity into useful reports. Connect the remote server first. Follow the [MCP setup guide](/docs/developers/mcp), then start a new chat with the TalkToHumans tools enabled. ## Find contacts Your next useful conversation may already be in your LinkedIn network. Describe the person you want. Your AI client can turn that into structured searches across the contacts you have synced or imported, rank the results, and save the shortlist as a TalkToHumans view. ### What the MCP can search The `linkedin_list_contacts` tool can search across: → name and LinkedIn URL → headline, role, about text, and experience → current or previous company → location → company size → TalkToHumans tags and notes → connected or unconnected relationship state → pending invitation state → contacts inside a specific conversation Your AI client can combine several searches, inspect the results, and keep paging when it needs a wider sample. ### Find leads Give the AI a real ICP and tell it how to rank the results. ```text theme={null} Use TalkToHumans MCP to find 20 people in my LinkedIn network who match this ICP: [Describe the company, role, size, geography, problem, and buying signal.] Return a table with: - name - current role and company - LinkedIn URL - why the person fits - relationship context - a 1 to 5 fit score - one useful outreach angle Do not draft messages yet. Let me review the shortlist first. ``` The last line matters. Research first. Outreach after you agree with the list. ### Find hires ```text theme={null} Search my TalkToHumans contacts for potential hires for this role: [Paste the role, required experience, location, and useful background.] Prioritize people I am already connected to or have spoken with before. Explain the evidence for each match and flag missing information. ``` This works well for warm sourcing, past candidates, former colleagues, and people whose job changed since you last spoke. ### Find partners and warm paths ```text theme={null} Search my LinkedIn network for people who could help me reach this audience: [Describe the audience.] Look for agencies, consultants, creators, investors, recruiters, operators, past customers, and people working with the companies I care about. Separate the results into: 1. direct prospects 2. potential partners or referrers 3. people who may provide a warm introduction ``` The strongest result is often someone who knows the buyer, not the buyer themselves. ### Save the shortlist in TalkToHumans Once the list looks right, ask: ```text theme={null} Create a TalkToHumans contact view called "[View name]" for the approved contacts. Add a short description explaining why this list exists. ``` The MCP uses `linkedin_create_contact_view` to create a focused list. Open it in TalkToHumans, review each person in context, and decide who deserves a message. ### Keep the judgment human AI is good at scanning profile data, finding patterns, and suggesting angles. It does not know every history, political detail, or reason to leave someone alone. Use it to narrow the network. You still choose the people and the message. ## Draft outreach Giving AI a LinkedIn send button would be fast and, honestly, a terrible idea. The MCP handles the useful preparation without sending anything. Give your AI client a set of profile URLs and the reason you want to contact them. It can import the contacts, organize them, add context, and save a draft for each person. You open TalkToHumans to review and act. That boundary is deliberate. ### What the workflow can do The TalkToHumans MCP can: → list the LinkedIn accounts you can access → import one contact from a `linkedin.com/in/` profile URL → enrich a contact and their primary company → add or remove tags → save account-scoped notes → create a contact view for the group → save a message draft for a contact or existing conversation The draft tool returns a TalkToHumans link and confirms that nothing was sent. The MCP does not currently apply a sequence or bulk-send the prepared messages. Use the app for that final step. ### Import a list and prepare drafts ```text theme={null} Use TalkToHumans MCP for the LinkedIn profile URLs below. Goal: [Explain the project and why these people were selected.] Account: [Name the LinkedIn account if the workspace has more than one.] For each valid profile: 1. import the contact 2. add the tag "[Tag name]" 3. add a note with the source and reason for outreach 4. draft a short, personal connection note or first message Then create a TalkToHumans contact view called "[View name]". Rules for the drafts: - use only facts present in the supplied context or contact profile - keep the tone low pressure - do not invent familiarity - do not send anything - return the TalkToHumans review link for every contact Profile URLs: [Paste the URLs] ``` The AI will call the import and draft tools once per person. If a profile fails, ask it to show the reason and continue with the rest. ### Draft from an existing conversation The MCP can also read a conversation before preparing the next reply. ```text theme={null} Find my conversation with [Name]. Read the recent messages and draft a reply. The reply should: - answer their last question directly - use the context already in the thread - stay under [length] - avoid making promises I did not give you Save the result as a TalkToHumans draft. Do not send it. ``` This uses `linkedin_list_conversations`, `linkedin_get_conversation_messages`, and `linkedin_create_message_draft`. ### Review and send from TalkToHumans The draft result includes an `app_url`. Open it to land on the right contact or conversation. Read the profile, tags, notes, company context, and any existing message history. Make sure the AI selected the right identity. Remove anything you would not have written yourself. Check names, claims, variables, links, and the reason for reaching out. For an unconnected person, use the draft as a connection-request note when appropriate. For an existing conversation, send or schedule it as a message. If the contact deserves a structured follow-up, apply a sequence in the app and tailor each step to that person. ### Why this is safer than direct AI sending Importing a profile, creating a view, adding notes, and saving a draft change TalkToHumans workspace data. They do not send a LinkedIn action. The final action stays inside the browser-based TalkToHumans flow: → you see the person and the draft → you choose whether to send → the extension checks the active LinkedIn identity → LinkedIn actions pass through the normal pacing and daily limits You get the useful part of AI, which is research and preparation, without handing it an invisible send loop. Importing a contact does not prove the person's data is current. Review the LinkedIn profile and any freshness warning before relying on the result. ## Team reports Your manager asks how LinkedIn is going. "Busy" is probably not the report they had in mind. TalkToHumans gives admins the useful numbers without exposing private conversation content. Ask a question through the MCP or API, compare teammates or accounts, and let your AI turn the result into a table or graph. There is no in-app analytics dashboard yet. Your AI client is the reporting interface. ### What you can measure → distinct people contacted → replies to outbound conversations → reply rate → connection requests sent and accepted → connection-request acceptance rate → activity by teammate and LinkedIn account → activity by channel, source, and event type → runs, replies, and rates for each template Filter by date range, account, user, event type, channel, or source. ### What reporting does not contain The activity ledger does not include message bodies, attachments, notes, drafts, or AI-generated copy. It tells a manager what happened without turning private DMs into surveillance data. ### How rates are counted Reply and acceptance rates use distinct contacted people and activated template runs. Re-syncing a message or receiving several replies in one conversation should not make the rate look better than it is. ### Run a weekly manager report Ask an authorized AI client: ```text theme={null} Use TalkToHumans MCP to prepare our LinkedIn manager report for the last 7 days. Show: - people contacted - replies and reply rate - connection requests sent - connection requests accepted and acceptance rate - results by teammate - results by LinkedIn account - the 5 best and 5 weakest templates with enough runs to be useful Call out missing attribution or unusually small samples. Do not include or infer message content. ``` ### Build a trend graph Ask the AI to call `activity_summary` once for each period you want to compare, then graph the results. ```text theme={null} Build a 6-week LinkedIn activity report from TalkToHumans. For each week, collect: - people contacted - reply rate - connection requests sent - connection-request acceptance rate Create: 1. a line chart for weekly volume 2. a line chart for reply and acceptance rates 3. a short written explanation of the biggest changes Keep zero values visible. Mark weeks with incomplete or unattributed data. ``` The client can create the chart in whatever format it supports, such as an inline graph, spreadsheet, notebook, or image. ### Compare teammates without rewarding spam Raw send volume is easy to game. Pair activity with outcomes and context. ```text theme={null} Compare LinkedIn activity by teammate for the last 30 days. Show volume, replies, reply rate, connection requests, and acceptance rate. Do not rank people on volume alone. Flag small samples and missing user attribution. Suggest questions a manager should ask before drawing a conclusion. ``` ### Compare templates ```text theme={null} Use TalkToHumans activity data to compare template performance over the last 30 days. Only compare templates with enough activated runs to make the result useful. Show runs started, replies, reply rate, connection requests, acceptances, and acceptance rate. Separate message templates from connection-request flows. ``` ### The MCP tools → `activity_list` for filtered activity records.
→ `activity_summary` for grouped metrics and rates. Both tools accept date, account, user, event-type, channel, and source filters. ### Report with the API Use `GET /v1/activity` for filtered records and `GET /v1/activity/summary` for grouped metrics. Admins manage workspace API keys.