Overhear runs an MCP server, so an AI assistant like Claude can read your mentions, mark them, and manage keywords and categories for you.
Connect
The server lives at /mcp on your Overhear address, for example https://your-overhear-host/mcp.
Clients that support OAuth, like Claude, only need the server URL. Add it as a custom connector. The client opens Overhear, you sign in and approve access, and the client gets its own token.
Open Account, create a key under API keys and copy it. Keys are shown once. Send it with every request as either header:
The assistant works within your organization's projects, like you do in the app.
Tools
| Tool | What it does |
|---|---|
list_projects |
Your projects with their keywords, description, lead profile, custom categories and webhook URL. Start here for project and keyword ids. |
list_mentions |
A project's mentions, newest first, with type, sentiment and urgency. Filter by source, category or urgency, show only unrelated ones or only saved testimonials. With leads: true, the project's leads instead, best fit first, with fit, fit_reason, intent, and source evidence. Fit measures customer suitability, not purchase probability. Returns 50 by default, at most 150. |
list_recent |
The newest related mentions and fitting leads across all your projects, each with its project, like the workspace overview. Returns 20 by default, at most 100. |
set_lead_profile |
Save buyer context: project_id, ideal_customer, problems, alternatives, buying_signals, and exclusions. All five fields are required (up to 1,000 characters each); empty strings clear them. Replaces the full profile and returns the saved values. Applies to future scoring; does not start searches. |
add_keyword |
Track a new keyword in a project. kind is brand (default) or lead; problem (a problem id) files a lead keyword under that problem; competitor: true tags a brand keyword naming a rival. It starts at the default pace: brand keywords on X and Reddit every hour, competitors and lead keywords every 6 hours, LinkedIn every 6 hours, YouTube, Hacker News and the web daily. sources limits it to some of x, reddit, linkedin, youtube, web and hackernews; omit it for all. |
list_problems |
A project's problems (one-line statements of what the product solves, the unit of lead finding), each with its leads in the last 30 days, its phrasings and whether each runs or was paused, plus any suggested phrasing. |
add_problem |
Add a problem to a project and start finding people who have it: project_id, statement, optional phrasings. Without phrasings Claude derives three (a pain, an intent and a question phrasing); each becomes a lead keyword at the default pace. Phrasings the project already tracks are skipped. Costs credits per phrasing on overhear.dev. |
list_moments |
A project's competitor moments: bursts of complaints about a tracked competitor (outage, price change, shutdown, acquisition, bad release) with a one-line summary, usual and peak complaints per 6 hours, start and end, and its leads. Pass a moment id to list_mentions as moment (with leads: true for its leads). |
set_keyword_cadence |
Set how often a keyword searches each source: keyword_id and cadence, hours per source id (1, 6 or 24), e.g. {"x": 1, "web": 24}. A source left out is off. |
draft_reply |
Draft a reply to a mention or lead, with a tracked link. Returns the text; nothing is posted. Costs 15 credits on overhear.dev. |
mark_replied |
Record that a reply was posted, or take it back. |
resume_keyword |
Resume a lead keyword Overhear paused after poor search quality (paused and pause_reason in list_projects). It is not paused again for 30 days. |
remove_keyword |
Stop tracking a keyword and delete its mentions. |
add_category |
Add a custom category with a name and a description. Up to 10 per project. |
remove_category |
Remove a custom category. Its mentions are sorted again on the next scan. |
save_testimonial |
Save a mention as a testimonial, or take it out. |
set_mention_relevance |
Mark a mention as about your product, or as unrelated (hidden from the feed). Overrides Claude's judgment. |
whoami |
Your account and how the request authenticated. |
echo |
Returns the message you send, to test the connection. |
In the browser
The dashboard also offers the same tools to an agent driving your browser, through the draft WebMCP standard (document.modelContext). Nothing to connect: every signed-in page registers them, calls run inside the tab you are signed in to, and they act as you.
The names and arguments are the ones above, minus whoami and echo, plus context: your organization, the project open on the page and every project's id and URL. Agents should call context first.
Browsers without WebMCP ignore the tools; the /mcp server is unaffected either way. In Chrome, inspect and call them under Application > WebMCP in DevTools.
Try it
For ready-made workflows built on these tools, install the agent skills.
Ask your assistant things like: