The FMP MCP server for ChatGPT requires the right surface and permissions
FMP publishes an official remote MCP endpoint that turns its API catalog into discoverable tools. ChatGPT can connect to remote MCP servers through custom apps, formerly called custom connectors. The important caveat is that ChatGPT availability depends on plan, user role, workspace policy and rollout—not only on whether the FMP URL works.
Business · Enterprise · Edu
OpenAI documents full MCP and developer mode on ChatGPT web for these workspace plans. Admin, owner, authorized-developer and RBAC rules differ by plan.
Pro
OpenAI currently documents developer-mode custom apps with read/fetch permissions for Pro, not full write support. Recheck the current product before setup.
Web and workflow limits
Custom MCP apps are documented as web-only. Agent mode does not use custom apps; deep research can use read/fetch actions only.
How to add the FMP MCP server to ChatGPT
Confirm plan, role and workspace policy
Identify whether you are a Business admin/owner, an Enterprise/Edu admin or authorized developer, or an eligible Pro user. If developer mode is absent, do not treat that as an FMP server failure.
Enable developer mode
Use the current Settings → Apps → Advanced Settings or workspace app-creation path documented for your plan. Enterprise/Edu access can be granted with RBAC; Business administration is more centralized.
Create a custom app
From user or workspace app settings, choose Create, enter clear metadata and provide the official FMP remote MCP endpoint. ChatGPT does not connect directly to a local MCP server.
Choose authentication and scan tools
FMP documents a URL with the API key in its query string. Treat the completed URL as a credential, choose the applicable authentication option, then use Scan Tools and review the discovered catalog.
Test as a draft
Use a low-risk, narrow read request. Inspect the proposed tool, symbol, dates, period, units, returned record and account usage. Do not publish merely because discovery succeeded.
Publish with controlled access
Only admins/owners publish workspace apps. Review safety warnings, configure permitted actions and user access where supported, then document owner, approved use cases and revocation procedure.
https://financialmodelingprep.com/mcp?apikey=YOUR_FMP_API_KEYNever paste a completed URL into screenshots, tickets, prompts, analytics or source control. The placeholder is documentation; the completed URL is a secret.
Primary sources checked: FMP official MCP server guide, OpenAI developer mode and MCP apps, and OpenAI Apps in ChatGPT.
Govern the app, the FMP key and the financial data separately
| Control plane | What to configure | Evidence |
|---|---|---|
| CHATGPT | Developer access, app owner, draft/published state, action controls, RBAC groups and tester access. | Current app settings and workspace policy. |
| FMP | Dedicated key, plan entitlement, quotas, endpoint access, commercial use and rotation. | Account dashboard, order form, usage and key inventory. |
| TOOLS | Discovered tool names, schemas, risk classification, expected arguments and approved use cases. | Tool scan snapshot and acceptance results. |
| DATA | Display, storage, derivation, redistribution, provenance, retention and deletion rules. | Applicable FMP agreement and product design record. |
OpenAI notes that newly published or changed actions require administrative review. Enterprise/Edu admins can refresh actions, see changes and control access; Business update behavior can require recreating and republishing an app. That makes tool-catalog changes a release-management event, not an invisible server-side detail.
Test the FMP app with a bounded research question
Expected tools only
Record server identity, scan time, tool names, schemas and any unexpected or duplicated action.
Explicit scope
Require symbol, statement type, period, date range, currency and maximum records; reject ambiguous requests.
Traceable facts
Keep tool, arguments, provider, retrieval time, record date, units and raw result separately from model interpretation.
Visible limits
Exercise invalid keys, unavailable endpoints, 429 responses, empty history and partial results without unbounded retries.
Separate ChatGPT access failures from FMP data failures
Developer mode is missing
Check plan, web surface, admin enablement, user role, RBAC and workspace policy. This happens before FMP is contacted.
Tool scan fails
Check the remote URL, key exposure or invalidity, endpoint reachability, authentication selection and server compatibility.
Tool call fails
Check FMP plan entitlement, required parameters, symbol coverage, rate limit, account status and the actual error payload.
Answer is wrong
Inspect the tool arguments and raw result. Recheck period, units, currency, adjustments, revisions and whether model interpretation introduced the error.
Only connect trusted MCP servers. OpenAI explicitly warns that unsafe or untrusted servers increase security risks, including prompt injection. Retrieved financial news, filings and text must be treated as untrusted content that cannot override application policy.
Use QVeris as a provider-aware governance layer
Policy before invocation
QVeris can help expose authorized financial tools with user, environment and task boundaries; require dates and symbols; cap result size; separate test and production credentials; and preserve source metadata for downstream answers.
Boundaries remain explicit
QVeris does not enable ChatGPT developer mode, publish workspace apps, create FMP entitlements, grant redistribution rights or guarantee financial accuracy. Those remain OpenAI, workspace, FMP, contract and human-review responsibilities.
Related QVeris guides: FMP MCP server for Claude, financial RAG, and SEC filings for AI agents.
FMP MCP server for ChatGPT questions
Can ChatGPT connect to the FMP MCP server?
Eligible ChatGPT users can configure remote MCP servers as custom apps. Access depends on plan, role, workspace policy, web availability and current rollout.
Which ChatGPT plans support custom MCP apps?
OpenAI currently documents full MCP for Business and Enterprise/Edu, while Pro developer mode is limited to read/fetch permissions. Verify current availability before rollout.
Does ChatGPT connect to a local FMP MCP server?
Not directly. OpenAI says ChatGPT connects to remote MCP servers. For private or local infrastructure, use a supported secure tunnel architecture rather than exposing it casually.
How do I add the FMP MCP server to ChatGPT?
Confirm eligibility, enable developer mode, create a custom app, provide the official FMP remote endpoint, choose authentication, scan tools, test a draft and publish under workspace policy.
Is the FMP API key safe in a custom app URL?
The completed URL is a credential that can leak through sharing, logs or screenshots. Use a dedicated key where possible, restrict visibility, monitor usage and rotate or revoke after exposure.
Can QVeris govern FMP tools in ChatGPT workflows?
QVeris can add authorized discovery, parameter rules, quotas, provenance and review gates. It does not create ChatGPT access, publish the app or grant FMP data rights.