Model Context Protocol: Any Data, Any AI, with FME

Welcome to the official Instrumentl Model Context Protocol (MCP) server. This server bridges your AI Assistants (Claude, ChatGPT) or other MCP clients directly to your Instrumentl account.

Instrumentl MCP connects your MCP client to your Instrumentl account, enabling your AI assistants to extract insights and take real actions in Instrumentl directly from a conversation.

Note: This is a distinct product surface from Instrumentl's API, which remains a feature exclusive to the Full Lifecycle, University, and Enterprise Plans.

Availability

The Instrumentl MCP server is available to all accounts with active subscriptions (regardless of plan) or active free trials.

Accounts in Maintenance Mode will not have access.

Note: Free trial accounts can access the MCP, but are limited to 15 records returned per project per session — specifically via the List Project Matches and Search Grants tools.

What is Instrumentl MCP?

Instrumentl MCP is the Model Context Protocol interface to Instrumentl. It lets your AI assistants take real actions in Instrumentl, such as surfacing matches for an Instrumentl project, creating tasks, and updating records.

Your AI assistant operates under the same governance model as the rest of Instrumentl: workspace-level access controls and user permissions. Instrumentl is SOC 2 Type II certified.

How Instrumentl MCP works

Instrumentl MCP has three parts:

Your MCP server. It holds the tools your AI can call. Each MCP client gets its own server, so you might have one for Claude and another for ChatGPT.
Your MCP client. Any application that supports MCP, such as Claude, ChatGPT, Copilot, Cursor, etc.
Tools. Each tool is a single action in a connected app, such as Show me Grant Matches for XYZ Project in Instrumentl.

When you ask your AI to do something, it picks the right tool, fills in the details from your conversation, and calls it. Instrumentl runs the action and returns the result.

Most servers use dynamic tool discovery, so your AI can find and enable the tools it needs during the conversation. The MCP tools section below covers the full scope of tools available via the Instrumentl MCP.

Data security & authentication

An Instrumentl MCP server connection links one MCP client (e.g., Claude or ChatGPT) to one Instrumentl account. The server hosts the actions your MCP client can call and routes each call through your existing Instrumentl account.

Instrumentl supports two MCP clients through direct integrations: Claude and ChatGPT. For both systems, you will sign in with OAuth. Claude or ChatGPT registers Instrumentl as an MCP server and sends you through an Instrumentl sign-in.

Once you're set up, you can describe what you want in a conversation and your AI handles the rest.

Get started

Instrumentl currently supports direct integrations with Claude and ChatGPT. Select your AI tool below to get set up:

Note: Support for additional MCP clients — including Copilot, Gemini, and others — is on our roadmap. Stay tuned for updates!

MCP tools

Below are all the tools your AI assistant can use to read, create, and update data in your Instrumentl account. The list will grow as new tools are added.

Title	Tool name	Type	Description	Example prompt
Get Account	`get_current_account`	Read	See which Instrumentl account you're connected to.	Which Instrumentl account am I signed into?
List Users	`list_users`	Read	See everyone on your team's account.	Show me all users in Instrumentl.
List Projects	`list_projects`	Read	See all the projects set up in your account.	What projects do I have set up for our food security programs?
List Project Matches	`list_project_matches`	Read	See recommended grant matches for a project.	For the literacy program, show recommended grant matches with deadlines before December 31.
List grants	`list_grants`	Read	See grants your team has saved, hidden, or added yourselves.	Show me the grants we uploaded ourselves.
List tracked grants	`list_saved_grants`	Read	See the grants in your Tracker and where each one stands in your pipeline — including grant status, amount requested and awarded, custom field values, and a direct link to the grant record in Instrumentl.	Show every grant we've been awarded this fiscal year for the housing project.
Save grant match	`save_match`	Write / Update	Add a recommended match to your Instrumentl Tracker.	Save that Ford Foundation match to our tracker.
Update tracked grant	`update_saved_grant`	Write / Update	Update details on a tracked grant — including status, owner, amounts requested and awarded, and grant period dates. *Note:* Custom fields are not updatable at this time.	We were awarded $50,000 by Hewlett for the period from July 1, 2026, to June 30, 2027. Update the tracker.
Hide saved grant	`hide_saved_grant`	Write / Update	Remove a grant from your Tracker. *Note:* You can restore it later.	We don't want to pursue the Bloomberg grant anymore — remove it from our tracker.
List funder profiles	`list_funders`	Read	See funder profiles your team has saved, hidden, or added.	Show me all the funders we've saved to our tracker.
List funder contacts	`list_funder_contacts`	Read	See the contacts your team has added to a funder's profile.	Who are our contacts at the Ford Foundation?
List upcoming tasks	`list_tasks`	Read	See upcoming tasks and custom deadlines.	What tasks do I have coming up this week?
Create or update tasks	`upsert_tasks`	Write / Update	Create or update tasks, including who owns them and when they're due.	Create a report task on our Kresge grant, assign it to Dana, due March 15.
List grant budget	`list_budgets`	Read	See the budget for an awarded grant.	Show me the budget for our Old National Bank Foundation grant.
List grant expenses	`list_expenses`	Read	See expenses logged against an awarded grant.	What have we spent so far on the Old National Bank Foundation grant?
Create expenses	`create_expenses`	Write / Update	Log new expenses against an awarded grant. *Note:* We recommend adding only planned expenses to maintain your accounting software as your source of truth.	Show the impact of a $12,000 planned expense on Feb 1 on our Old National Bank budget.
Ignore expenses	`bulk_update_expenses`	Write / Update	Exclude an expense from your budget totals without deleting it. The amount, date, category, and description stay on record.	Ignore that duplicate expense so it doesn't count against our budget.
Delete expenses	`bulk_destroy_expenses`	Write / Update	Permanently delete expenses. *Note:* This cannot be undone.	Delete the three test expenses I added while learning the system.
Submit MCP feedback	`submit_feedback`	Write / Update	Send feedback directly to the Instrumentl team about the MCP.	I want to send feedback to the Instrumentl team about the Claude connector.

Chaining endpoints together

Below are some natural questions that the AI agent can answer by chaining multiple Instrumentl endpoints. You can use these to spark ideas for 'skills' you want to build for your AI agent (Claude, ChatGPT, e.g.).

Use case	Endpoints chained	Example prompt	How it works
Spenddown risk — what's falling behind	`list_saved_grants` → `list_budgets` → `list_expenses`	Which awarded grants are underspending with fewer than 3 months left in the grant period?	Agent finds saved grants with status “awarded” Agent reviews budget and expenses for awarded grants (to compare BvA) Agent computes burn rate against time left. Agent surfaces awarded grants underspending with 3 months left in the grant period "Underspending" is agent-computed, not a native flag.
Discovery → action — find and save	`list_project_matches` → `save_match`	Find new grants for our housing program and add the three best fits to our tracker.	Agent finds housing program project Agent reads recommended grant matches Agent picks 3 strong fit grants and saves them to the user’s tracker
Cross-project duplication — are we doubling up	`List_saved_grants`, `grant_id`	Are we actively pursuing the same grants across multiple projects? Flag conflicting statuses or ask amounts.	Agent groups the Tracker by grant_id (only saved grants, not hidden) Flags grants saved across multiple projects Flags divergent status/amount.
Funder coverage — who has something open	`list_grants` → `list_funders`	Which funders we're tracking have grants open with deadlines before December 31?	Agent reviews open grants and filters to grants with deadlines before December 31, 2026 (list_grants) Agent resolves the funders associated with grants and lists those funder names (list_funders)
Deadline hygiene — backfill report tasks	`list_saved_grants` → `list_tasks` → `upsert_tasks`	For every awarded grant, is there a report task? Create one wherever it's missing and ask me to slot in a deadline.	Agent finds saved grants with status “awarded” Agent checks existing tasks Agent creates the missing report tasks and prompts user for deadline
Team workload — who's overloaded	`list_tasks` → `list_users`	Show open tasks by teammate, and flag anyone with more than five past-due tasks.	Agent views open tasks grouped by assignee Agent names users with 6+ open tasks flagged as “past-due” via list_users.
Pipeline report — where do we stand	`list_saved_grants`	Pipeline summary for our housing project: grants per stage, total requested vs. awarded this year.	Agent groups the tracker by grant statuses with sum requested per stage Agent shows sum requested vs. awarded this year
Category reallocation — single grant	`list_budgets` → `list_expenses`	On the Kresge grant, which budget lines are underspent with two months left?	Agent finds awarded grant requested Agent reviews grant budget and actuals per line Agent computes the remaining balance and surfaces which budget lines are underspent, with 2 months remaining in the grant period. No native "underspent" field.

Limitations

Deadlines

Application cycles & fiscal years

When your connect AI agent lists upcoming deadlines or sorts grants saved to your tracker (all projects) by deadline, it surfaces the next available cycle for each grant — not the cycle you've indicated for that saved opportunity. This is because the AI agent cannot read the "Your Cycle" data or the selected FY in your saved grant. Cross-reference any deadline list with your tracker to confirm which opportunities you're actively pursuing.

In the example below, an AI agent would list this saved grant with an August 1, 2026, deadline, because that's the upcoming deadline for the current cycle. However, in the Tracker, this awarded grant is saved to FY 2024 and was for the Aug 1, 2024, cycle.

As mentioned above, connected AI agents do not have access to the cycle or fiscal year data when getting saved grants from your account. When the AI agent reads your account's saved grants, it sees the official grant page info for each opportunity saved from the Instrumentl database (think: 'Funding Opportunity' tab of a saved grant) as well as select data from the 'Overview' tab, such as amount requested, amount awarded, and grant period.

As a workaround, be sure to prompt your AI agent to include the status of the saved grants you're asking about, as this will help you determine whether it's a future or past cycle/deadline (e.g., Awarded = past; Planned = future).

Custom grants

AI agents can see custom grants, including their status. However, the custom deadline entered is not visible to the connected agent.

FAQs

Can I connect to multiple AI tools — for example, both Claude and ChatGPT?

Yes, there are no limits on how many MCP clients you can connect to your Instrumentl account.

Can I connect multiple Instrumentl accounts to a single AI tool?

It depends on the tool. With ChatGPT, you can connect multiple Instrumentl accounts simultaneously — from the Instrumentl plugin page, under "Connected accounts," select + Connect another account.

If you name the account in your request (e.g., "Find matches for the food program in my Social Services of Minnesota account"), ChatGPT will use it directly; otherwise, it will ask you to clarify if the context is ambiguous.

I'm a consultant with multiple client profiles. Can my AI tell them apart?

Not yet — client profiles aren't exposed through the MCP today, but this is on our roadmap.

Will the AI provider train on my data?

This depends on your plan with the AI provider, not Instrumentl. Check your settings with the AI tool you're using. As a general rule: business and enterprise tiers typically exclude your data from training by contract, while personal and consumer tiers may use conversations for training unless you opt out in your privacy settings.

What data can the AI provider actually access?

Only what's exposed through the MCP tools — scoped to your own permissions in Instrumentl. The connection can't reach data you don't already have access to.

Contact us

If you have any questions about the Instrumentl MCP server or need help getting set up, we're here for you! Message our friendly support team via the chat bubble when logged into your account, or email us at hello@instrumentl.com.