Connect Your AI Assistant: Agent Access Setup

What Agent Access does

Most of us keep an AI assistant open all day. It is sharp about general things and blank about the one thing you actually need: what is moving in federal health IT this week, and whether it is worth your time.

Agent Access closes that gap. You connect your assistant once, and from then on it can read your current Mission Meets Tech intelligence: market intel, open opportunities, and the pipeline you track on the site. You ask it what to chase, and it answers from live data, with the sourcing. It is read-only, so it can look but never change anything.

Before you start

You will need two things:

• An MMT Premium membership with Agent Access turned on. Agent Access is a premium add-on. If you open the connection page and see a prompt to "Add AI access to your plan," that add-on is not on your account yet. You can add it from the pricing page, or reply to your account contact.
• An AI assistant that accepts a connector. The two most common are Claude Desktop and ChatGPT. Any assistant or tool that lets you set a custom authorization header works too. If your company runs a locked-down corporate version, see Troubleshooting.

Step-by-step setup

What to ask once it is connected

A few prompts to get your team started:

• "What should I be watching in federal health IT this week?"
• "Show me open DHA solicitations under NAICS 541512."
• "Summarize everything in my pipeline and flag anything with a deadline in the next 30 days."
• "Who is the likely incumbent on this recompete, and when does it expire?"
• "Is this opportunity worth our time given what we do? Give me the sourcing."

What the market-entry reference gives your assistant

Connections created on or after September 20, 2026 can carry four more permissions: Market-entry reference, State Medicaid and state procurement, Agency org charts and, off by default, Signal Chain. The reference layer is the same one behind the Premium Market Entry and State Medicaid pages, served read-only:

• Buyers. VA, DHA, CMS, ASTP/ONC, NIH, CDC, HRSA, IHS, ARPA-H, ASPR/BARDA and the state Medicaid segment, with the routes, authorization paths and vehicles that apply to each.
• Vehicles. Every IDIQ and GWAC in the MMT tracker with an ordering status (open, closing soon, pre-award, closed, cancelled), the ordering end date and the date the dataset was generated.
• Authorization paths. FedRAMP Rev5 and 20x, CMS Rapid Cloud Review, DoD impact levels, VA, ONC certification, GovRAMP, TX-RAMP, MARS-E.
• State Medicaid. All 56 agencies, the 90/10 and 75/25 funding rules, APD thresholds, certification, cooperative purchasing routes and the dated demand signals.
• Innovation doors and compliance rules. SBIR Phase III, ARPA-H, BARDA, the DHA CSO, MTEC, VA Pathfinder, CMS Innovation Center models; FAR 3.4, FAR 9.5, the Lobbying Disclosure Act, the Procurement Integrity Act and the Byrd Amendment.

Every answer carries the date the record was verified and the time it was retrieved, so your assistant can say "as of" on every claim. A connection made before September 20, 2026 does not have these permissions; revoke it and create a new one to add them.

What every record promises

Every record your assistant reads carries the same five fields, so it can tell you how much to trust a claim without a second call:

• source_url. The primary public source, never an MMT page. When the record is MMT's own output, such as a score, it is null and the gap list says where the provenance lives.
• retrieved_at. When I last read the source.
• confidence. verified means the primary source was read inside the record's freshness window. reported means a secondary account or a medium-confidence read. stale means the read is older than the window: opportunities one day, vehicle ordering status and state procurement records seven days, org charts thirty days, statutory conditions ninety days, my hand-maintained Contract Tracker forty-five days. Stale means check before relying on it, and a good assistant says so instead of answering from it.
• as_of. When the fact itself is effective, where that differs from the read.
• gap. What I do not have, as a list of field and reason. A gap is never filled with a guess.

State Medicaid procurement coverage

State coverage is honest about its edges. Ask for coverage first (the tool is mmt_states_coverage, or GET /api/v1/states/coverage) and you get, for every state and territory, whether each entity is live, partial or not covered: the agency and its procurement portal, solicitations on file, the module landscape (who runs claims, pharmacy, EVV and the rest), cooperative vehicles such as the NASPO ValuePoint MES portfolios, participating addenda, and the CMS conditions for enhanced funding. Full detail exists for the ten highest-spending states. A question about a state I do not cover for that entity returns a coverage gap with the coverage row, never an empty list your assistant might read as "nothing there."

Request ids, attribution and your usage statement

Every response carries an X-Request-Id header, and every error body a request_id; if something looks wrong, send me that id and I can find the exact call. Your assistant can also send an X-MMT-Client-Ref header on each call. It is opaque to me: I never read it, only group by it, so your monthly usage statement can show which of your clients or projects the calls belonged to. The statement lives under each connection's Usage button on the AI integrations page: calls against the monthly allowance, anything over it priced at the published per-call rate, and the per-client breakdown. The next section has the numbers and where a connection pauses.

Calls, the monthly allowance and what happens past it

Each connection includes 5,000 calls a month. Only calls that return data count. An error, a refused call or a rate limit never uses up your allowance and is never billed.

Past the allowance it keeps working at $0.01 a call, added to your Agent Access subscription's next invoice, for up to 5,000 more calls ($50.00). Then it pauses until the first of the next month, so a runaway assistant cannot run up a bill you did not see coming. I email you at 80 percent of the allowance, on the first extra call, and if it pauses. Reply to any of those and I will set a different limit for that connection. Billing for extra calls starts with October 2026 usage; nothing before that is charged.

Extra calls are billed to the Agent Access add-on, so they need it on your account. A connection without the add-on pauses at the allowance until the next month.

The Usage button on each connection shows the month so far: calls that counted, errors that did not, where the limit sits, and the split by client if your assistant sends a client ref. What you see there is exactly what is billed.

Running my tools through your assistant

With the right permissions your assistant can also run the paid tools for you: Ask MMT, Pursuit Score, Compliance Check and, if you turned it on, Signal Chain. They run exactly as they do on the site, against your own monthly allowances, and Ask MMT returns the same sources list the web widget shows, untouched.

Your data and privacy

Best practices

• One key per person. If your team shares a membership, each person creates their own connection rather than passing one key around. It keeps the audit log clean and lets you revoke one person without disrupting everyone.
• Name connections clearly. "Trish - Claude Desktop" beats "connection 1" when you have a few of them.
• Use the 90-day expiry. It renews often enough to stay secure without becoming a chore. Avoid "Never."
• Treat the key like a password. Do not paste it into email or chat in plain text. Copy it, use it, and if it ever gets loose, revoke it and make a new one.
• Revoke when someone leaves or changes roles. It takes one click and takes effect instantly.
• Check the Activity log now and then to see what is being read and confirm everything looks right.

Troubleshooting

I see "Add AI access to your plan"

That means your membership does not have the Agent Access add-on yet. Add it from the pricing page, or reply to your account contact and we will turn it on.

My assistant will not connect

Double-check that you pasted the full key, that the header reads exactly Authorization: Bearer followed by your key, and that the base URL is https://missionmeetstech.com/api/v1.

My company runs a locked-down corporate assistant

Some enterprise deployments block outside connectors by default. If that is your situation, your IT team may need to allow the connector. Give them the base URL above and let them know it is an outbound, read-only HTTPS connection authenticated with a bearer token. If they want more detail, point them to the connection page, which links to the full technical reference.

I lost my key

No problem. Revoke that connection and create a new one. Keys are shown once on purpose, so a fresh one is the fix.

Get help

Stuck on any step? Email support@missionmeetstech.com and we will walk you through it. For questions about adding Agent Access to your plan or setting up a team, see the pricing page or reply to any premium email.