For the complete documentation index, see llms.txt. For the full corpus, see llms-full.txt. For raw markdown, append .md to this URL or send Accept: text/markdown.

Use a Coding Agent

Drive SessionSight from any MCP-compatible coding agent

Last updated: May 21, 2026

SessionSight ships an MCP server that exposes ~170 tools so a coding agent can do real work in your account: track goals, run split tests, manage flags, query analytics, and wire any of it directly into your code.

If your editor’s agent speaks MCP, you can pair it once via OAuth and drive SessionSight from inside the conversation. No copy-pasting between dashboards and code.

Install

Claude Code

bash

/plugin marketplace add SessionSight/sdks
/plugin install sessionsight@sessionsight-sdks

The plugin auto-registers the MCP server. Authentication happens on your first tool call: your browser opens to a SessionSight consent screen where you choose the company (and optionally the specific properties) the agent can touch.

Cursor, Windsurf, Zed, and other MCP clients

Point your client’s MCP config at the SessionSight server:

text

https://api.sessionsight.com/mcp

The transport is Streamable HTTP. Most MCP-compatible clients accept a URL and handle the OAuth dance automatically. Refer to your editor’s MCP setup docs for the exact config syntax.

Use It

Ask in plain language. The agent picks the right tool (or chain of tools):

text

what are my goals?
create a split test for the new pricing page
summarize last week's campaigns and which earned signups
create a goal for revenue and wire it into checkout

When you ask the agent to wire a goal, split test, or feature flag into your code, the response includes the SDK install command, the init snippet, and the call-site snippet. The agent drops them in for you at the right place.

Reads run concurrently

You can ask multi-part questions ("what are my top funnels and which split tests are running?") and the agent will issue the read calls in parallel.

What the Server Can’t Do

DELETE operations are rejected at the MCP layer. To delete goals, properties, campaigns, or anything else, use the dashboard. This is a deliberate guardrail so an agent can’t destroy data without you opening the UI.

Property Scope

When you authorize the agent, you choose which properties it can touch. The default is “any property in this company” so day-to-day flows just work. Picking “specific properties only” locks the install to a frozen subset. Re-pair from your dashboard’s Authorized Clients page to widen, narrow, or revoke access. Existing tokens die immediately on revoke.

Privacy and Security

Every action is attributed in your audit log with source: "mcp", the tool name, and the client identifier.
The server never reads or returns secret API keys. Server SDKs’ secrets are only retrievable from the dashboard’s API Keys page.
The pair is bound to your company plus property selection from consent. A pair scoped to one property can never query another, even if you have access to both.

Troubleshooting

Tool calls fail with “MCP token doesn’t have analytics:read scope”

The bearer was minted before the latest scope augmentation. Re-pair: in Claude Code, delete ~/.config/sessionsight/auth.json and re-trigger any tool call. Other clients: clear the cached MCP credentials and re-authorize.

Browser opens to “Log in to SessionSight before continuing”

You’re not signed in to the dashboard. Open SessionSight in another tab, sign in, then re-trigger the agent’s pair command.

Tool calls fail with “session expired” or 404

The upstream MCP session expired (one-hour TTL). Trigger any tool call again and the client will re-initialize automatically.