# Connect Kindling to your AI assistant (MCP)

> Docs · Connectors · Canonical page: https://kindlinghq.com/docs/mcp/

Kindling runs an MCP server, so you can connect your workspace to Claude, ChatGPT, Cursor, or another MCP client and ask questions about your advocacy program. The connector is read-only and is included on every plan.

**Connector URL:** `https://app.kindlinghq.com/api/mcp`

Paste this into your assistant's connector settings and sign in with your Kindling account. There is no API key to create.

Read-only. No API key. Included on every plan.

## On this page

1. [Before you start](#1-before-you-start)
2. [Connect the server](#2-connect-the-server)
3. [Client-specific steps](#3-client-specific-steps)
4. [Tool reference](#4-tool-reference)
5. [Response conventions](#5-response-conventions)
6. [Data access](#6-data-access)
7. [Security](#7-security)
8. [Troubleshooting](#8-troubleshooting)
9. [Disconnect](#9-disconnect)
10. [FAQ](#10-faq)

## 1. Before you start

MCP (Model Context Protocol) is an open standard for connecting AI assistants to software. Kindling hosts an MCP server at the URL above. Once you connect it, your assistant can read your program data and answer questions about it.

- **Account type.** You need an active member account in an active company workspace. Advocate (creator) accounts cannot connect yet.
- **Role.** Owner, admin, manager, and viewer can all connect. The connector shows the same data your role already sees in the dashboard.
- **Plan.** Included on all plans. Nothing to enable or purchase.
- **Client.** Any client that supports remote MCP servers over streamable HTTP. Claude, ChatGPT, and Cursor are tested before each release.

The connector is read-only. It cannot approve submissions, edit programs, send payouts, or invite members. Those actions stay in the dashboard.

## 2. Connect the server

1. **Add the connector URL.** In your assistant's connector or MCP settings, add `https://app.kindlinghq.com/api/mcp` as a custom connector.
2. **Sign in.** A Kindling sign-in window opens. Use the same credentials you use for the dashboard. If your company uses single sign-on, it works the same way here.
3. **Approve read access.** Grant the read-only permission. Access is scoped to your account and your workspace.
4. **Verify.** Ask your assistant to summarize your program performance for a specific date range, then compare the numbers against the same range in your dashboard. They should match.

## 3. Client-specific steps

- **Claude:** Settings → Connectors → Add custom connector. Paste the URL and sign in. Works in the web app and the desktop app.
- **ChatGPT:** Settings → Connectors → Add. Requires a plan that supports custom connectors.
- **Cursor:** Settings → MCP → Add new server. Use the same URL; sign-in opens in your browser.

Other MCP clients work as well. The server supports dynamic client registration, so a client we have not pre-registered can still connect.

## 4. Tool reference

Eleven read-only tools. Your assistant selects the right one based on your question. You can also name a tool directly if you want a specific result.

| Tool | Returns | Parameters |
| --- | --- | --- |
| `get_account_overview` | Workspace name and status, your role, program counts by status, enrolled advocate count, and lifetime approved submissions, impressions, and spend. | None |
| `list_programs` | Programs with status, allowed platforms, posting rules, CPM rate, caps, payout frequency, and permitted reward types. | status, limit, offset |
| `get_program` | One program in full: description, guidelines, example post, platform rules, payout configuration, reward menu, and enrollment and submission counts by status. | programId or slug |
| `get_analytics_overview` | Impressions, spend, submission counts, approval rate, active advocates, and effective CPM. Defaults to the last 90 days. | from, to, programId |
| `get_program_performance` | All programs side by side: submissions, approvals, approval rate, impressions, spend, effective CPM against target CPM, and enrolled advocates. | from, to |
| `get_emv_report` | Total earned media value, average EMV per post, EMV-to-spend ratio, total engagements, and the EMV configuration used (CPM benchmark and engagement weights). | from, to, programId |
| `list_advocates` | Enrolled advocates with enrollment status, audience size and tags, and per-advocate approved posts, impressions, earnings, and last post date. | programId, enrollmentStatus, sort, from, to, limit, offset |
| `get_advocate` | One advocate: profile, audience information, program enrollments with status, totals including engagement, and their five most recent posts. | advocateId |
| `list_submissions` | Submissions with platform, post URL, reported impressions, likes, comments and shares, review status, rejection reason, and calculated payout. | programId, advocateId, status, platform, from, to, sort, limit, offset |
| `get_payout_summary` | Payout totals grouped by month and status (pending, processing, completed, failed), with a breakdown by reward type. Defaults to the last six months plus the current month. | from, to |
| `get_monthly_trends` | Monthly series of approved impressions, spend, and submission counts, oldest first, including months with no activity. | months (default 6, max 24), programId |

### Example questions

These map cleanly onto the tools above and are a reasonable place to start:

- Summarize our advocacy program performance this quarter.
- Who are our top ten advocates by impressions? By earnings?
- Which posts drove the most impressions last month, and on which platforms?
- How many submissions are waiting on review?
- What is our effective CPM, and what is our EMV-to-spend ratio?
- How much have we paid out this quarter, and how much is pending?
- Compare our programs. Which one is performing best?
- Which advocates have not posted in the last sixty days?
- Have impressions grown month over month this year?

## 5. Response conventions

- **Money** is returned as a decimal amount with an explicit currency of USD.
- **Dates** are ISO 8601 in UTC.
- **Date ranges** are echoed back in every filtered response, including the default when you did not specify one. If your assistant reports a number, you can ask it which range it applied.
- **Lists** return 25 results by default and 100 at most, along with the total count. Your assistant can page through the rest.
- **Terminology** is advocate, not creator or influencer, in every response.

## 6. Data access

The connector exposes what your dashboard already shows you, minus compliance-sensitive records. It does not widen access. A viewer using the connector sees what a viewer sees when they sign in.

**Available**

- Program configuration, rules, and guidelines
- Submission post URLs and reported metrics
- Advocate names, emails, social URLs, and audience information
- Payout amounts and statuses
- Earned media value figures and configuration

**Excluded**

- Tax and compliance data: EIN, payer address, W-9 status, 1099 filings
- Stripe and Tremendous identifiers
- Internal tokens and system identifiers
- Audit logs and notification contents
- Any other company's data

**Advocate personal data.** Names, emails, and social profiles appear in tool results because they appear on your advocates page. Tax records, banking details, and reward preferences are excluded at the field level and covered by tests.

## 7. Security

- **Authentication.** OAuth 2.1 against your existing Kindling account. No API keys are created, stored, or rotated. Tokens are scoped read-only.
- **Validation.** Tokens are checked on every request, so revoking access takes effect on the next call.
- **Tenant scope.** Derived from your signed-in account. Tools do not accept a company identifier as an input. A request for a record in another workspace returns the same not-found response as a record that does not exist.
- **Rate limit.** 60 tool calls per minute per workspace.
- **Audit.** Every tool call is written to your workspace security audit log with the member, tool, and timestamp.
- **Monitoring.** Errors and latency are tracked per tool, with sensitive fields scrubbed before they reach our monitoring tools.

## 8. Troubleshooting

| Message | What to do |
| --- | --- |
| This connector is for Kindling company accounts. | You signed in with an advocate account. Sign in with your company workspace account. Advocate access is not available yet. |
| No Kindling workspace found for this account. | Sign in with the email address you use for Kindling. A different email on the same domain will not resolve to your workspace. |
| Your account access is suspended or pending. | Contact your workspace admin to activate your membership. |
| Your workspace is not active. | Finish onboarding, or contact support@kindlinghq.com. |
| Rate limit reached for your workspace. | Wait a moment and retry. The limit is 60 calls per minute across everyone in your workspace. |
| Numbers do not match the dashboard. | Ask your assistant which date range it applied. Most differences come from a default range (90 days for analytics, all time for program performance). If the ranges match and the numbers do not, send both to support@kindlinghq.com. |

## 9. Disconnect

Remove the connector in your assistant's settings, then revoke the grant in your Kindling account settings. The next call fails authentication. No program data changes and nothing is deleted.

## 10. FAQ

**Does it cost extra?**
No. It is included on every plan.

**Can it change anything in my account?**
No. All eleven tools are read-only.

**Is my data used to train a model?**
Kindling does not send your data anywhere for training. What your AI provider does with a conversation is governed by your agreement with them, so check the data controls on your plan with that provider.

**Can my advocates connect it?**
Not yet. This version serves company workspaces. An advocate-side connector is on the roadmap.

**Can I connect more than one assistant?**
Yes. There is no cap on connected clients per member.

**Will write actions be added?**
Approving submissions and similar actions are planned, with an explicit confirmation step before anything is written.

**Where do I report a problem?**
support@kindlinghq.com. Include the question you asked, the tool your assistant used if it says, and the dashboard view you compared it against.