MCP help

Remote MCP server reference and troubleshooting

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI tools talk to outside tools and data sources. SeanPropApp ships an MCP server so the same 18 methodology modules you can run in the workspace also run from inside your AI tool, using your own subscription instead of an API key.

The remote MCP server (this page) is an OAuth-gated HTTPS endpoint. You add the URL once in your AI tool, sign in through your browser, and the modules appear as tools and prompts. Your content never touches our servers; we only assemble the methodology prompt and return it for your tool to execute.

Which AI Connection?

We strongly recommend running your analyses in the Browser app, where you hold a structured, guided discussion with the AI for each module. You can power the Browser app two ways: the Browser Command Line Interface (CLI) bridge, which lets you use your existing Claude or ChatGPT subscription, or Bring Your Own Key (BYOK), which is simpler to set up but bills token usage to your AI provider. You can also run the methodology directly inside your own AI tool through the Model Context Protocol (MCP). Pick whichever matches where you want to work.

Browser CLI bridge

RecommendedNo API key

Best when: You are comfortable setting up a command-line tool and want to power the full Browser app (sidebar, guided runs, export) with your existing Claude or ChatGPT subscription, with no separate per-token spend.

You need: Node on your machine. Run npx @seanpropapp/cli connect, then pair. Your content goes browser to localhost to the claude CLI; it never reaches our servers.

Install the bridge

Bring Your Own Key (BYOK)

Uses your API key

Best when: You want the Browser app with no local install, and are fine funding token usage directly with your AI provider, separate from any monthly subscription you may already pay for.

You need: An Anthropic (Claude) or OpenAI API key, stored encrypted in your browser only.

Add a key in Settings

In-host MCP

No API key

Best when: You would rather invoke the methodology modules directly inside your own AI chat or agentic workflow (Claude, ChatGPT, Cursor, or Claude Code).

You need: An MCP-capable AI tool. Add a URL (remote) or install the @seanpropapp/mcp npm package (stdio).

Set up MCP

What AI tools work with this MCP server?

The remote MCP server works with any MCP-capable client that supports OAuth-gated remote servers. Current known-working tools:

Claude (web, mobile, desktop, and Claude Code)
Cursor
ChatGPT Pro (via custom connectors)
Any other MCP-capable client that supports add-by-URL

The local stdio path uses the @seanpropapp/mcp npm package with a personal access token. It works in any stdio-capable MCP host including Claude Desktop, Cursor, and Claude Code (local install).

Adding SeanPropApp to your AI tool

Step-by-step setup, the MCP URL, and per-tool configs (Claude, ChatGPT, Cursor, and others) live on the setup page:

Open the SeanPropApp MCP setup page

Troubleshooting

Every error response from the SeanPropApp MCP server includes (a) what you saw, (b) why it happened, (c) how to fix it, and (d) where to get more help. The ten classes below are the complete list; if your host shows something outside this list, treat it as an unexpected 500 and email support with the request ID.

401 INVALID JWT

Your AI host shows: "Your sign-in to SeanPropApp expired. Please reconnect."

Why: SeanPropApp access tokens are short-lived for security. When the token expires, every subsequent tool call from your host returns 401.
How: Reconnect from your host. In Claude, open the connector settings, find SeanPropApp, and click "Reconnect". Your host will walk you through the sign-in flow again. No data is lost; your previous outputs stay in your Drive.
Where: If reconnection itself fails, see "503 Sign-in provider temporarily unavailable" below.

401 WRONG AUDIENCE

Your AI host shows: "Token issued for a different SeanPropApp environment. Please reconnect."

Why: You signed in to a different environment (for example, a preview deploy or a local dev server) and your host cached that token. The production server rejects it.
How: Remove the connector entry in your host, then add it again using the production URL: https://prop.seanoneill.com/api/mcp/handshake. Sign in fresh when prompted.
Where: If you intentionally use a non-production environment, make sure the URL in your host config matches the environment you signed in to. Email support@seanoneill.com if you need a list of valid environment URLs.

400 UNKNOWN MODULE_ID

Your host shows: "Unknown module '{module_id}'. Available: {list}. Try 'list_modules' first."

Why: The module name you (or the model) passed is not in the registry. Common causes: a typo, a deprecated module name, or a module that was renamed between methodology versions.
How: Ask your host to call the list_modules tool first. It returns the current set of module IDs. Then re-issue your request using one of those IDs.
Where: SeanPropApp keeps the registry stable across patch versions and bumps METHODOLOGY_VERSION on any rename. If you suspect a module was removed, check the methodology page for the current list.

400 MISSING REQUIRED FIELD

Your host shows: "Missing required field '{field}'. Required: module_id, company. Optional: persona, prior_outputs, business_unit, initiative_name, company_url, investor_questions."

Why: Every module needs at minimum a module_id and a company name. The model omitted one of these.
How: Restate your request with the missing field. For example: "Run the moat module on Stripe." If the model keeps forgetting a field, prompt it explicitly: "Pass company='Stripe' to the run_module tool."
Where: For complex analyses, you can pre-fill defaults by using the start_analysis prompt template (Claude shows it as a slash command).

400 MODULE DEPENDENCY MISSING

Your host shows: "Module '{module_id}' requires output from '{dep_id}'. Run '{dep_id}' first, then call '{module_id}' with prior_outputs={...}."

Why: Several SeanPropApp modules build on earlier ones. The moat module reads the competitive landscape; unit economics reads the ICP. Skipping a dependency starves the downstream module of context.
How: Run the dependency first. Then re-call the target module and pass the dependency output in prior_outputs. A good host will do this automatically when prompted: "Run the dependency, then the module."
Where: Start with the SETUP prompt template if you are running a new company end-to-end. It sequences the foundation modules in the right order. See the methodology page for the full dependency graph.

413 PAYLOAD TOO LARGE

Your host shows: "Field '{field}' exceeds {limit}KB. Trim and retry."

Why: One of the fields you sent (most often prior_outputs or investor_questions) exceeded the per-field size cap. The cap exists to keep latency predictable and to fit within the 60 second serverless timeout.
How: Trim the largest field. For prior_outputs, pass only the modules the target module actually depends on (not every prior module). For investor_questions, summarize before sending.
Where: If you genuinely need larger payloads (for example, a multi-business-unit analysis), split the run into smaller scoped requests. The methodology supports this via the business_unit field.

429 RATE LIMITED

Your host shows a per-tier message and a Retry-After header. Free: "Daily limit reached for the free tier. Upgrade or try again tomorrow." Beta: "Burst limit reached. Wait {N}s and retry." VIP / admin: only hit on cascade storms; same retry pattern.

Why: SeanPropApp caps requests per user per tier to keep the service responsive for everyone. The cap is per-token, per-hour, and per-day.
How: Wait the seconds shown in the Retry-After header. Your host typically waits automatically and resumes. If you hit limits frequently, request a tier upgrade.
Where: Tier upgrades: email support@seanoneill.com with your account email and the use case. Beta and VIP tiers are granted manually.

503 SEANPROPAPP TEMPORARILY UNAVAILABLE

Your host shows: "SeanPropApp is temporarily unavailable. Try again in {Retry-After}s."

Why: An upstream dependency (database, edge cache) is unhealthy. SeanPropApp fails fast rather than serving stale or partial responses.
How: Your host will retry automatically using the Retry-After value. If retries also fail, wait a minute and try again manually.
Where: Check the status page or recent updates if outages persist past 5 minutes. Email support@seanoneill.com if your host shows this error continuously for more than 10 minutes.

503 SIGN-IN PROVIDER TEMPORARILY UNAVAILABLE

Your host shows: "Sign-in provider is temporarily unavailable. Existing sessions continue; new sign-ins will retry."

Why: SeanPropApp uses WorkOS for the OAuth handshake. If WorkOS is degraded, new sign-ins fail even though existing tokens keep working.
How: If you already have a valid token, keep working; only new connections are affected. If you need to sign in, wait a few minutes and try again. WorkOS recovery is usually fast.
Where: If sign-in still fails after 15 minutes, email support@seanoneill.com so we can confirm the outage and post an update.

500 UNEXPECTED

Your host shows: "Something went wrong. Request ID: {requestId}. Email support@seanoneill.com if this persists."

Why: An unhandled server error. Every 500 response includes a correlation ID so we can trace exactly which request failed and why.
How: Retry once. If it fails again, copy the full Request ID from the error message and email it to support@seanoneill.com. We can trace the failure end-to-end in our logs without seeing your content.
Where: support@seanoneill.com. Include the Request ID, the rough timestamp, and (if comfortable) the module you were running. We never log your prompts or outputs; the correlation ID alone is enough.

Migrating from the stdio (npm) version

If you already use the @seanpropapp/mcp stdio package, nothing changes. Both paths are co-equal and fully supported: the stdio version keeps working and you can run both at once. The remote server is recommended for new connections because it works on mobile and the web, requires no local install, and updates automatically when SeanPropApp ships new modules.

To switch, add the remote URL in your AI tool (see “Adding SeanPropApp to your AI tool” above), confirm it works on a test run, then remove the stdio entry from your tool config. There is no data migration; tokens are independent across the two surfaces, and analysis content lives in your browser or Drive, not in the MCP server.

Methodology updates

When SeanPropApp ships a new module or revises an existing one, the change is available to MCP users within 24 hours. The methodology version response is edge-cached with a 24 hour TTL; on the next refresh, your host picks up the new module list and the new prompts automatically.

To see the change immediately, restart your host or remove and re-add the connector. The METHODOLOGY_VERSION field on every response tells you which version of the methodology served your prompt.

Privacy and security

The methodology stays on our server. Your content (company name, prior outputs, investor questions, generated module text) is never logged, never stored, and never used to train any model. The server assembles a prompt from your inputs and returns it to your host; the host runs inference against your subscription. We see only structural facts: which module ran, which host called it, request size, latency, and HTTP status.

Authentication uses OAuth 2.1 with WorkOS as the identity provider. WorkOS issues short-lived access tokens to your AI tool; our server does not store refresh tokens. Tokens are audience-scoped to this environment. See the privacy page for the full data flow.

Contact support

Stuck, confused, or saw something this page does not explain? Email support@seanoneill.com. Include the correlation ID (request ID) from any error message and the rough timestamp; we can trace the failure end-to-end without seeing your content.