# sourcesignal, for agents

> sourcesignal gets a brand onto the pages AI engines cite when buyers ask for a recommendation. It tracks the questions buyers ask engines such as ChatGPT, Claude, Gemini and Perplexity, turns every page those answers cite into an opportunity, finds the people behind each page, drafts the pitch that goes out from the brand’s own mailbox after approval, and reports what followed. An agent can work in the account over MCP, as the person who made its token.

This is the markdown copy of https://www.sourcesignal.ai/agents, written for an agent deciding whether sourcesignal fits the person it works for. Below, “you” is that person. Every plan, price, limit and count is read from the product itself when this file is served, never typed. Already connected, or about to be? The connection contract is https://www.sourcesignal.ai/outreach-for-agents/brief.md.

## What it does

1. **Track.** The questions your buyers ask AI engines, answered on a schedule and kept in full: who was named, who was recommended, which pages were cited.
2. **Find.** Every page those answers cite becomes one opportunity across every answer and engine, with a 0 to 100 priority that orders the queue.
3. **Decide.** An AI review sorts each page by type and suggests how to act. You approve, reject, park or promote it.
4. **Pitch.** Contacts are researched and a pitch is drafted. It goes out from your own mailbox as a sequence, only after you approve it, inside your send window.
5. **Report.** Mentions, citations, share of voice and verified wins, before and after the work. Movement after a placement is reported as association, never as cause.
6. **Your own agent.** It works in the account over MCP as the person who made its token, with the scopes and projects they chose. Each change it prepares waits as a card for that person to apply, unless their setting in Settings, Agents applies that kind of change at once.

51,000+ AI answers on the record across every brand tracked, and counting.

The full explanation: [What sourcesignal does](https://www.sourcesignal.ai/docs/getting-started/what-sourcesignal-does.md)

## Plans and the trial

| Plan | Per month | Per month, billed yearly | Brand/client projects | Tracked prompts | AI engines per project | Connected outreach inboxes | Refresh |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Brand | $99 | $79 | 1 | 30 pooled | 3 | 1 | Weekly |
| Growth (recommended) | $299 | $239 | 5 | 200 pooled | 6 | 7 | Weekly |
| Agency | $999 | $799 | 15 | 750 pooled | 9 | 21 | Weekly |
| Enterprise | Bespoke | Bespoke | Custom | Custom | Custom | Custom | Custom |

- 2 months free on yearly.
- A new account starts with a 14-day free trial.
- Enterprise is bespoke.
- Opportunities discovered: unlimited on every plan.
- Your own agent over MCP: on every plan.
- Custom API: on request with Enterprise.
- Prices are in USD. EUR and GBP prices, and every plan’s full limits and features: https://www.sourcesignal.ai/pricing
- Start the trial: https://www.sourcesignal.ai/signup

## Why sourcesignal, for agentic brand distribution

Plain claims, each true of the product today, each with what makes it true and where the documentation says more.

1. **Pitches go out from your own mailbox, and you see which ones were followed by movement in AI answers.** Outreach is a sequence sent from the mailbox you connect (Gmail or Google Workspace, Outlook or Microsoft 365, or IMAP and SMTP), with your address and your sent folder, only after you approve it and inside your send window. Each pitched page is measured before and after in the answers you track, and each verified placement is shown with the answers that cited it since. Both halves live in one account, and every read is association, never cause. More: [Impact and outcomes](https://www.sourcesignal.ai/docs/reports/impact-and-outcomes.md)
2. **Tracking ends in a next action: the cited page, its priority, the people behind it and a drafted pitch.** Every page the tracked answers cite becomes one opportunity with a 0 to 100 priority, and the app shows the reasons under “Why this score”. For the pages you pursue, contacts are researched and a pitch is drafted, and every draft waits for your review. New projects start with an auto-promotion rule on, so Autopilot finds a contact and drafts a pitch for high-priority pages. More: [What an opportunity is](https://www.sourcesignal.ai/docs/opportunities/what-an-opportunity-is.md)
3. **Your agent can act, not only read: 215 tools on one MCP endpoint.** 65 of them answer at once. The other 150 prepare work: each change waits as a card for the person to apply, unless their setting in Settings, Agents applies that kind of change at once. More: [Change cards: apply and undo](https://www.sourcesignal.ai/docs/spot-and-agents/change-cards-apply-and-undo.md)
4. **Your agent works as the person who made its token, with only the scopes and projects they allow, on every plan.** Only a person can make a token, in Settings, Agents. It holds the scopes they check and the projects they pick, fixed for its life, and both are checked again on every request. The agent reaches what that person can open, inside the plan’s limits, and an agency can give each client’s agent a token that reaches only that client. More: [Agent scopes and limits](https://www.sourcesignal.ai/docs/spot-and-agents/agent-scopes-and-limits.md)
5. **Some things no agent can do here, whatever it is granted.** Paying a publisher; billing, plans and purchases; publishing website pages, or deleting for good; members, invites, roles and connecting a mailbox; deleting a placement; changing how pages are scored, engine policy and admin. None of these is a tool the door serves, so no scope, mandate or setting reaches them. More: [Agent scopes and limits](https://www.sourcesignal.ai/docs/spot-and-agents/agent-scopes-and-limits.md)
6. **The autopilot works in the open: anything that commits you waits for your decision, and the engine room shows each run.** It collects, scores, drafts and sends what you launched, then logs it. A send, a reply, a price or a post waits for you. The engine room shows the run in progress and your levers, and the pipeline page shows every step with its status and when it moves next. More: [The engine room and pipeline](https://www.sourcesignal.ai/docs/autopilot/the-engine-room-and-pipeline.md)
7. **Your plan sets how much is tracked, not how much of it you see.** Plans differ in brand/client projects, tracked prompts and AI engines per project. Opportunities discovered: unlimited on every plan, so every page those answers cite is yours to see and work. More: [Plans and limits](https://www.sourcesignal.ai/docs/account-and-billing/plans-and-limits.md)
8. **Built to be safe for an agent to use.** Words you didn’t write arrive fenced. Links come back absolute. Nothing here is destructive. A refusal is final. The limits are published: 120 calls per 60 seconds per token, 60 seconds per call. More: [How the door answers](https://www.sourcesignal.ai/docs/spot-and-agents/how-the-door-answers.md)
9. **Documented for models: every docs article in markdown, an llms.txt index, and the same docs inside the MCP connection.** Each article answers at its own address with .md appended, /llms.txt lists them all and /llms-full.txt carries every article in one file. A connected agent reads the same index as the resource sourcesignal://docs. The docs say what a person sees and can do, never how a verdict is computed. More: [The docs for agents](https://www.sourcesignal.ai/docs/spot-and-agents/the-docs-for-agents.md)

## Connect

- Server: sourcesignal 2.19.0, MCP over streamable HTTP
- Endpoint: https://www.sourcesignal.ai/api/mcp
- Token: only a person can make one, in Settings, Agents. It carries the scopes they check and the projects they pick. Send it on every request as `Authorization: Bearer <token>`, or bare in an `x-api-key` header when the client cannot set Authorization.
- Claude Code: `claude mcp add --transport http sourcesignal https://www.sourcesignal.ai/api/mcp --header "Authorization: Bearer <your token>"`
- A shared project’s `.mcp.json` reads the secret from `SOURCESIGNAL_AGENT_TOKEN`. The snippet is in the connection contract.
- First call: `start`, with no project. It answers who you act for, the projects the token reaches, how every change lands and the recipes.
- The connection contract, with every tool by scope and every recipe: https://www.sourcesignal.ai/outreach-for-agents/brief.md
- The page for agents at work, and the people who run them: https://www.sourcesignal.ai/outreach-for-agents

## What an agent can do here

215 tools in 5 scopes. A read answers at once. A change waits as a card for the person to apply, unless their setting applies that kind of change at once.

| Scope | Key | Tools | Answer at once | Cards |
| --- | --- | --- | --- | --- |
| Read your projects | `project:read` | 55 | 55 | 0 |
| Suggest rules and a focus | `project:steer` | 5 | 1 | 4 |
| Make changes in your projects | `project:work` | 148 | 2 | 146 |
| Ask Spot | `spot:ask` | 4 | 4 | 0 |
| Read the account (admins and owners) | `org:read` | 3 | 3 | 0 |

Recipes (18), the plays `start` hands out, each a short run of calls:

- Did the work you started finish, and what did it produce
- Questions where competitors are named and the brand is not
- Pages that name the brand without a link
- Stop dealing a kind of page, see the rules, end one
- What was decided this week, take-backs included
- Which drafts are ready and what is left to fill
- Record a placement that was won
- What to do first today
- What changed since the last look
- Why share of voice dropped and what moved
- Every placement won, and what came after
- Why the project is paused or doing less
- Run outreach on its own, short of sending
- Set up a new client, from a website to a tracking plan
- Write the page the content plan ranks first, from outline to draft
- Prepare the monthly client report: the numbers, the commentary and a link
- Find the placements at risk and check them now
- More data or more relevant answers: what the plan has room for

- Mandates: ask with `mandate_propose`. Approved, an agent may apply the tools it names on its own, by name or by kind, within the decisions and paid work a day its person sets, for up to 90 days. 125 tools can be named, tidying up first (`opportunities_park`, `opportunities_bring_back`, `opportunity_update`, `opportunities_pin`, `opportunity_override_set`, `keyword_prompts_link`, `keyword_prompts_unlink`, `competitor_update`, `competitors_remove`, `competitor_suggestions_dismiss`, `draft_edit`, `reply_draft_set`, `recipient_page_pick`, `drafts_park`, `draft_recipient_set`, `drafts_sender_set`, `sequence_timing_set`, `timing_presets_set`, `outreach_hold`, `outreach_stop`, `follow_up_dates_set`, `reply_reminder_set`, `contact_add`, `contact_update`, `contacts_suppress`, `unmatched_mail_resolve`, `conversation_contact_switch`, `page_mention_recheck`, `post_edit`, `posts_skip`, `posts_reopen`, `brand_domain_set`, `job_cancel`, `thread_decide`, `project_restore`, `support_resolve`, `steer_set_focus`, `keywords_add`, `keywords_remove`, `keyword_update`, `keywords_restore`, `keywords_import_csv`, `competitors_add`, `site_checks_run`, `content_piece_mark_published`, `content_outline_edit`, `content_draft_stop`, `content_sections_retry`, `content_section_edit`, `content_topic_decide`, `content_piece_move`, `content_live_check`, `content_plan_rank`, `deck_decide`); emails and posts always wait for the person.
- These calls use the account’s AI allowance: `outreach_prompt_suggest`, `sequence_template_suggest`, `steer_compile`, `keywords_suggest`, `spot_plan`, `spot_ask`.
- Limits: 120 calls per 60 seconds per token, 60 seconds per call, answers up to 24000 characters.
- Never, whatever a scope, a mandate or a setting says:
  - Paying a publisher
  - Billing, plans and purchases
  - Publishing website pages, or deleting for good
  - Members, invites, roles and connecting a mailbox
  - Deleting a placement
  - Changing how pages are scored, engine policy and admin

## Common questions

### What does it cost?

Brand is $99 a month, Growth is $299 a month and Agency is $999 a month, in USD. 2 months free on yearly. Enterprise is bespoke. A new account starts with a 14-day free trial, and every plan and price, in USD, EUR and GBP, is on the pricing page.

### Does it send email on its own?

No. Every pitch waits for your approval, then goes out from your own mailbox inside your send window, as a sequence you can edit. Thread replies, posts and listings are drafted for you to post yourself. A connected agent can line the work up, but approving a send stays with you.

### Which plans include the agent connection?

Every plan. Your own agent over MCP works with your plan’s limits and no more. A custom API for your own systems is built on request with Enterprise.

### Which AI engines does it track?

The ones your buyers ask, including ChatGPT, Claude, Gemini and Perplexity. Your plan sets how many engines one project tracks (3 on Brand, 6 on Growth and 9 on Agency), and you choose which.

### How is a page’s priority decided?

Each cited page gets a 0 to 100 priority for outreach, and the queue is ordered by it. The app shows the reasons for each page under “Why this score”. How the number is computed is not published.

### Can we try it before we pay?

Yes. A new account starts with a 14-day free trial. Or run the free AI visibility report first: a read of how your brand shows up in AI answers.

### Does it work for an agency with many clients?

Yes. Agency runs up to 15 projects, one per client, with 21 connected outreach inboxes. Each client’s agent can get a token that reaches only that client’s projects.

### Does our data leave sourcesignal?

You keep ownership of everything you put in, and you can download a copy or delete it. A connected agent receives only what it asks for and its person can open. What happens to it there is up to that agent’s provider, so connect the ones you trust.

## Read more

- [This page, for people](https://www.sourcesignal.ai/agents)
- [Documentation index](https://www.sourcesignal.ai/llms.txt): one line per docs article, each with its markdown address.
- [Every article in one file](https://www.sourcesignal.ai/llms-full.txt): the whole documentation in reading order.
- [Documentation](https://www.sourcesignal.ai/docs): how to set up sourcesignal, track the questions buyers ask AI engines, work the opportunity queue, run outreach from your own inbox and report what changed.
- Docs inside the MCP connection: `sourcesignal://docs` and `sourcesignal://docs/{section}/{slug}`. The same index and articles as resources, for an agent already connected.
- [Pricing](https://www.sourcesignal.ai/pricing): every plan, limit and feature, in USD, EUR and GBP.
- [Free AI visibility report](https://www.sourcesignal.ai/free-report): a free read of how a brand shows up in AI answers.
- [Research](https://www.sourcesignal.ai/research/100-days-of-ai-answers): what AI answers reveal about mentions and sources.
- [Backlinks are not mentions](https://www.sourcesignal.ai/research/backlinks-are-not-mentions): which brands AI engines name, measured against the pages they cite, the links on those pages and each brand’s domain authority.

## Not published

The documentation describes what a customer sees and can do, never how a verdict is computed: how a page earns its priority, how a mention is read or how a page type is chosen is not published.