PostHog MCP vs Aident Loadout for Claude Code and Codex

Use PostHog's official AI plugin when PostHog is the center of the job and you want its first-party tools, skills, and OAuth flow inside Claude Code or Codex. Use Aident Loadout when the same agent must combine PostHog with other services behind one connection, discovery, preflight, and audit layer. The choice is not about which product can query PostHog. Both can. It is about whether you need a focused PostHog surface or a shared integration control plane.

The Short Comparison

Those differences make the decision fairly clean. Choose the official plugin for depth inside PostHog. Choose Loadout for breadth and shared controls across a larger agent workflow.

What the Official PostHog Plugin Gives You

PostHog's official plugin packages its hosted MCP server together with client-specific installation and more than 30 skills. Those skills teach the agent PostHog-specific workflows such as HogQL query patterns, experiment lifecycle management, feature-flag audits, error investigation, data-warehouse troubleshooting, and SDK instrumentation.

For Claude Code, the official installation is:

claude plugin install posthog

Then open Claude Code, run /mcp, select the PostHog server, and complete the browser OAuth flow.

For Codex, PostHog documents this path:

codex plugin marketplace add PostHog/ai-plugin
codex

Inside Codex, run /plugins, select PostHog, and install it. PostHog also offers a wizard for installing the MCP server into supported clients:

pnpm dlx @posthog/wizard mcp add

This is the strongest default when an agent's main responsibility is to inspect analytics, debug errors, manage flags, or run experiments in PostHog. The plugin carries first-party product vocabulary and current PostHog practices with it.

What Aident Loadout Changes

Loadout puts PostHog inside a shared Action layer. The agent discovers an Action by intent, inspects its current schema, preflights the exact input, and executes it through a connection held in Aident Vault. The same workflow applies when the next step moves from PostHog to GitHub, Slack, Lark, a CRM, a data warehouse, or another connected service.

Install Loadout by giving your agent this instruction:

Follow https://aident.ai/SETUP.md

Then verify the account and Vault before discovery:

aident account auth status
aident vault vault --action status

aident capabilities search \
  --query "PostHog list projects and run a read-only HogQL query" \
  --types '["action"]' \
  --targetEnv staging

Copy the exact public Action name returned by search into capabilities get, capabilities preflight, and capabilities execute. Do not guess an identifier or reuse an old schema from a prompt.

The reviewed staging catalog on August 12, 2026 exposed more than 500 PostHog Actions. Representative reads covered organization and project discovery, ad hoc queries, dashboards, feature flags, experiments, and warehouse metadata. A larger generated surface is useful for cross-service composition, but it is not automatically easier to use than PostHog's smaller first-party tool set. Search by a precise job and inspect only the Actions you need.

Run the Same Read-Only Smoke Test

Whichever path you choose, test it with a narrow read before granting the agent a write task.

Ask the agent to:

1. list the PostHog projects it can access;

2. select one project by name and report its numeric project ID;

3. count $pageview events and distinct visitors for one exact path over the last seven complete days;

4. return the query, time window, and result without saving an insight or changing a dashboard; and

5. stop if the project is ambiguous, the query expands beyond the named path, or the tool requires a write.

With Loadout, inspect the selected query Action and preflight the exact project ID and query body first. In the August 12 check, both project discovery and the reviewed ad hoc query returned zero-credit preflights. That is an observation, not a permanent price guarantee. Recheck the quote whenever the input or Action version changes.

With the official PostHog plugin, confirm the selected project and time range in the response. PostHog says connecting to its MCP server and calling MCP tools is free, but some tools use LLMs internally and may be billed as PostHog AI spend. Keep that separate from ordinary analytics queries.

Where Each Option Wins

Choose the PostHog plugin for a PostHog-native operating loop

A product engineer investigating a conversion drop might inspect the funnel, open the largest error group, review a feature flag, and propose an experiment. All four steps live inside PostHog, so the first-party tools and skills are the shortest path. The plugin also reduces setup drift because PostHog owns the MCP endpoint, OAuth flow, product vocabulary, and bundled instructions.

Choose Loadout for a cross-service evidence loop

A growth agent might query a PostHog funnel, enrich the affected accounts in a CRM, open a GitHub issue, and send a reviewed summary to Slack or Lark. PostHog supplies the behavioral evidence, but it is only one system in the job. Loadout keeps discovery, credentials, preflight, and Action history consistent as the workflow crosses provider boundaries.

This is also useful when the same authenticated capability must be available to more than one supported agent host. The connection lives in Vault instead of being copied into every prompt or local configuration.

Security and Write Boundaries

Do not paste a PostHog personal API key into a prompt, repository, shell history, or reusable skill. Use PostHog OAuth for the official MCP path or a reviewed Vault connection for Loadout.

Treat a POST endpoint as a write until the exact contract proves it only evaluates a query. Some APIs use POST for read-only query execution because the query body is too complex for a URL. Conversely, a friendly natural-language prompt can still create or change a feature flag, experiment, dashboard, cohort, or assignment rule.

Before any consequential action, require the agent to state:

• the selected PostHog project;

• the exact object it will create or change;

• the fields and values that will change;

• whether the operation can affect users or production traffic;

• the live price or credit quote; and

• the read-back that will prove the intended result.

Start with read-only access where possible. Add writes one use case at a time instead of granting a broad surface because it is convenient.

A Practical Selection Rule

Pick the official PostHog plugin if all three statements are true:

• PostHog is the primary system for the job;

• the agent benefits from first-party PostHog skills; and

• client-specific plugin installation is acceptable.

Pick Aident Loadout if any of these statements is true:

• the workflow must cross PostHog and at least one other service;

• the team wants one Vault and Action audit boundary across agents;

• exact preflight pricing is part of the execution policy; or

• the agent should discover live schemas instead of depending on a hand-maintained integration prompt.

You can also use both without duplicating credentials in prompts. Keep the official plugin for deep PostHog work and use Loadout for workflows whose evidence or consequences cross into other services. Give each path a clear responsibility so the agent does not choose between overlapping tools at random.

Connect PostHog Through One Reviewed Path

Follow https://aident.ai/SETUP.md

Set up Aident Loadout and inspect one PostHog read-only Action

For the general Loadout workflow, read How to Use Aident Loadout. For credential boundaries, see Give AI Agents API Access Without Exposing Keys. After execution, verify attribution with the Aident Action usage audit.

Sources

• PostHog AI plugin repository, reviewed August 12, 2026.

• PostHog MCP documentation, reviewed August 12, 2026.

• PostHog API documentation, reviewed August 12, 2026.

• Aident Loadout, reviewed August 12, 2026.

• Live Aident Loadout staging catalog, schemas, preflights, and read-only PostHog query, inspected August 12, 2026.

Refresh this comparison when PostHog changes its plugin installation, MCP tool or skill inventory, OAuth or billing behavior, or when the Aident PostHog Action contract, Vault flow, or preflight policy changes.