Use Bridget from your AI assistant

Connect Claude, ChatGPT, Cursor or Codex to Bridget. Ask in your assistant’s chat, and Bridget does the work in your Teamroom projects.

Streamable HTTP
OAuth sign-in
No local install
Agent text ↓

Essentials#

Read before connecting.

What the Bridget MCP server is#

Bridget is an AI research agent in Teamroom, the platform that hosts your projects. The Bridget MCP server connects Bridget to any assistant that supports the Model Context Protocol (MCP), such as Claude, Claude Code, ChatGPT, Cursor and Codex. You ask in your assistant's chat. The assistant passes the request to Bridget, waits for the result and brings it back to you.

Through MCP you get the same Bridget as in Teamroom and over email. It runs internal and external primary research, drafts interview guides, runs deep research, builds data tables and slide decks, and synthesizes findings with evidence. Bridget saves everything it produces to your Teamroom project.

Bridget always works inside one Teamroom project, and you can use only the projects you are a member of. Each project's compliance template and your permissions still apply.

Server settings#

Most clients need only the server URL. Sign-in happens in your browser the first time you connect.

Setting	Value
Server URL	`https://api.bridgetownresearch.com/kratos-api/mcp`
Transport	Streamable HTTP (remote server)
Sign-in	Your Teamroom account, in the browser (OAuth)
Local install	None
Permissions requested	Read and create projects, read and send conversation messages, answer Bridget’s questions, and read Bridget’s outputs.

Before you start#

A Teamroom account. You sign in with it when you connect. If you don't have one, ask your organization's Teamroom admin to invite you.
Project membership. Bridget can work only in projects you are a member of. You can also create new projects through MCP.
Credits. Work Bridget does through MCP uses the same credits as in Teamroom.
A client that supports remote MCP servers with OAuth. Some clients make custom connectors available only on certain plans, or need an admin to enable them. Check your client's settings.

Connect your client#

Follow the steps for your client, then check the connection.

Every client follows the same pattern: add a server with the URL below, sign in with your Teamroom account when the client asks, then mention Bridget or Teamroom in your prompts so the assistant uses the Bridget tools.

text

https://api.bridgetownresearch.com/kratos-api/mcp

Choose your client#

Open Settings → Connectors and choose Add custom connector.
Name it Bridget and paste the server URL. Select Add.
Select Connect on the new connector. A browser window opens. Sign in with your Teamroom account and approve access.
In a chat, open the tools menu and check that Bridget is turned on.

On Team and Enterprise plans, an owner may need to add the connector for the organization first. Members then connect their own accounts. When Bridget has a question, Claude asks you in the chat.

Check the connection#

Ask your assistant:

prompt

Which Teamroom projects can I use with Bridget?

The assistant should call bridget_list_teamrooms and list your projects. If it answers without calling a tool, or says it has no access, see Troubleshooting.

Working with Bridget#

What happens after you ask.

How a request runs#

You ask. Mention Bridget or Teamroom and the project, for example “Ask Bridget in [project name] to draft an interview guide on churn.” The assistant sends the request with bridget_send_message.
Bridget works. Quick answers come straight back. Longer work, such as deep research or a deck, runs in Teamroom while the assistant keeps checking for the result.
Bridget asks. When Bridget needs a choice, an approval or details, such as launching a study or spending credits, the assistant shows you the question. Some clients show a form, and others ask in the chat. The assistant passes your answer back to Bridget.
You get the result. The assistant shows Bridget's answer. It can open notes and deliverables in full, and gives you a download link for files such as decks.

Long work keeps running in Teamroom even if your chat closes. Later, ask the assistant to pick up the conversation, or open it in Teamroom.

Projects and conversations#

Projects. A project (also called a teamroom) is a shared workspace in Teamroom. It holds uploaded files, interview studies, transcripts, notes, data tables, decks, members and a compliance template. Refer to projects by name; you never need an id. Ask the assistant which projects you have access to, and it lists them. To create one, ask for a new project and give it a name. Bridget asks for any missing setup, such as project type or stage, rather than guessing.

Conversations. Each request runs in a Bridget conversation inside the project, with shared context: earlier messages, files and results. The first message starts a conversation and gives it a friendly name, and later messages in the same chat continue it. To pick up earlier work, ask the assistant to list your Bridget conversations in the project and continue one. It can replay the transcript to catch up.

Documents#

To share documents with Bridget, ask your assistant. It gives you a secure upload link. Open the link, sign in to Teamroom and upload the files, and the assistant carries on. Upload links stay valid for 7 days.

For one request: the files go with that message and stay in that conversation.
For the project: Bridget adds the files to the project as indexed, citable sources for every future conversation, interview and deep research run.
For a new project: you upload the files while the project is being created.

Accepted formats are PDF, Word, PowerPoint, Excel, CSV and images. Always upload documents through the link. Bridget can then search them, cite them, and use them for interviews and deep research.

Example prompts#

Task	Prompt
Find your projects	“Which Teamroom projects can I use with Bridget?”
Create a project	“Create a new Teamroom project called [name] for a commercial due diligence on [company]. I have background documents to add.”
Check saved context	“What does Bridget know about the [project name] teamroom?”
Add project files	“Add my market reports to the [project name] teamroom so Bridget can use them.”
Interview guide	“Ask Bridget in [project name] to draft an interview guide for CFOs on procurement pain points.”
External interviews	“Have Bridget recruit 15 IT directors in the US for interviews on [topic], using the guide we agreed.”
Deep research	“Ask Bridget to run deep research on the European cold-chain logistics market, and show me the plan first.”
Data table	“Have Bridget build a table of the top 30 [sector] companies in the UK with revenue, headcount and funding.”
Synthesis	“Ask Bridget to synthesize the completed interviews in [project name], with supporting quotes.”
Slides	“Ask Bridget to turn the synthesis into a 10-slide deck for the steering committee. Send the storyline first.”
Resume work	“List my Bridget conversations in [project name] and continue the pricing one. What was the next step?”

Tool reference#

The 15 tools the server exposes. Your assistant chooses them; you don’t call them yourself.

Each tool shows what it's for, its input, its output and an example. Inputs are optional unless marked required. Projects and conversations are passed by name. Values in angle brackets, such as <project name>, are placeholders. New tools may be added over time, and your client always shows the current list.

The conversation result#

Three tools return Bridget's reply in the same format: bridget_send_message, bridget_await_response and bridget_respond_to_prompt. Every result starts with a status and the conversation's friendly name.

Status	Meaning and contents
`completed`	Bridget finished. Includes its answer, the sources it cited and any suggested next steps.
`running`	Bridget is still working. Includes recent progress, such as the steps it has completed. Call bridget_await_response to keep waiting.
`suspended`	Bridget has started a long task that runs in the background, such as a deep research run. Call bridget_await_response later to pick up the result.
`input_needed`	Bridget has a question. Includes a summary of what it is asking you to approve, such as a research plan, and then each question field with its id and options. Answer with bridget_respond_to_prompt.
`failed`	Something went wrong. Includes the error message.
`cancelled`	The response was stopped.

When Bridget produced notes or files, the result also lists them as artifacts. Each one has a name you can pass to bridget_read_artifact, plus a download link and its expiry time for files such as decks and data table exports.

Projects#

bridget_list_teamrooms#

What it’s for

Seeing which Teamroom projects you can use with Bridget. The assistant also uses it when a project name is missing or matches more than one project.

Input

None.

Output

A list of your project names. If you have no projects, it says so.

Example

You ask: “Which Teamroom projects can I use with Bridget?”

json

{}

bridget_create_teamroom#

What it’s for

Creating a new Teamroom project. Bridget asks for any setup details you didn't give, such as project type or stage, and never guesses them from the name.

Input

Field	Type	Description
name	string	Project name, at least 3 characters.
type	string	Project type.
otherType	string	Required when type is Others.
stage	string	Project stage.
background	string	Project background or research instructions.
hasSupportingDocuments	boolean	You want to add documents. Bridget waits for your upload.
uploadTicketId	string	The upload ticket, once your documents have finished uploading.
userAskedBridgetToDecide	boolean	True only when you asked Bridget to choose the missing details.

Output

If details are missing, Bridget’s question with the options to choose from. Once everything is set, a confirmation that the project was created. If you uploaded documents, it also lists the files being added.

Example

You ask: “Create a new Teamroom project called <project name>.” Bridget then asks for the project type and stage.

json

{ "name": "<project name>" }

bridget_get_teamroom_context#

What it’s for

Reading what Bridget has saved about a project: its background context, its glossary of terms, or both.

Input

Field	Type	Description
teamroom	string	Project name.
type	context \| glossary \| all	What to read. Defaults to all.

Output

The saved context and glossary text. If nothing is saved yet, it says so.

Example

You ask: “What does Bridget know about the <project name> teamroom?”

json

{ "teamroom": "<project name>" }

Conversations#

bridget_send_message#

What it’s for

Asking Bridget to do something. This is the main tool. Without a conversation name it starts a new conversation, and with one it continues that conversation. It waits about a minute for Bridget’s reply.

Input

Field	Type	Description
messageRequired	string	Your request.
teamroom	string	Project name, passed on every message.
conversation	string	The conversation to continue. Omit it to start a new one.
uploadTicketId	string	Attaches documents you uploaded to this message. Each upload can be attached once.

Output

A conversation result. For quick requests it is usually completed. For longer work it is running or input_needed.

Example

You ask: “Ask Bridget in <project name> to draft an interview guide for CFOs on procurement pain points.”

json

{
  "teamroom": "<project name>",
  "message": "Draft an interview guide for CFOs on procurement pain points."
}

bridget_start_conversation#

What it’s for

Opening a new, empty conversation when you specifically ask for one. For a normal request, bridget_send_message starts a conversation on its own.

Input

Field	Type	Description
teamroom	string	Project name.
conversation	string	A friendly name for it, such as “market sizing chat”.

Output

The friendly name of the new conversation. Later messages use this name.

Example

You ask: “Start a new Bridget chat in <project name> called market sizing chat.”

json

{ "teamroom": "<project name>", "conversation": "market sizing chat" }

bridget_list_conversations#

What it’s for

Finding earlier Bridget conversations in a project, so you can pick one up again.

Input

Field	Type	Description
teamroom	string	Project name.
limit	integer, 1–50	How many to show. Defaults to 20.

Output

A list of conversations by friendly name. If there are none, it says so.

Example

You ask: “What Bridget conversations do I have in <project name>?”

json

{ "teamroom": "<project name>" }

bridget_continue_conversation#

What it’s for

Picking up an earlier conversation, so new messages continue it with all its context.

Input

Field	Type	Description
teamroom	string	Project name.
conversation	string	The conversation's friendly name.

Output

A confirmation naming the active conversation.

Example

You ask: “Continue the market sizing chat in <project name>.”

json

{ "teamroom": "<project name>", "conversation": "market sizing chat" }

bridget_get_transcript#

What it’s for

Catching up on a conversation: what was asked, what Bridget said and what it produced. Useful after you reconnect or pick up an old conversation.

Input

Field	Type	Description
conversation	string	The conversation's friendly name. Can be omitted when there is only one.
limit	integer	How many recent entries to show. Defaults to 100.

Output

The conversation in order: your messages, Bridget’s replies, and the artifacts it produced, each with a link.

Example

You ask: “What did Bridget conclude in the market sizing chat?”

json

{ "conversation": "market sizing chat" }

Waiting and answering#

bridget_await_response#

What it’s for

Waiting for Bridget to finish after a result said it is still working. Each call waits up to about 45 seconds. The assistant repeats it until Bridget answers or asks a question.

Input

Field	Type	Description
conversation	string	The conversation's friendly name. Can be omitted when there is only one.

Output

A conversation result with the latest status.

Example

A deep research request came back running, so the assistant keeps watching:

json

{ "conversation": "market sizing chat" }

bridget_respond_to_prompt#

What it’s for

Answering Bridget’s question, such as a clarification, an approval to launch, or a choice between options.

Input

Field	Type	Description
answerRequired	object	Your answers, keyed by the field ids in Bridget's question.
conversation	string	The conversation's friendly name. Can be omitted when there is only one.
answeredBy	string	Who answered.

Output

A conversation result as Bridget carries on with your answer.

Example

Bridget sends a research plan and asks for approval, and you reply “Approved, go ahead.” The field id comes from Bridget’s question; approval here is only an example.

json

{ "conversation": "market sizing chat", "answer": { "approval": "Approve" } }

bridget_cancel_response#

What it’s for

Stopping what Bridget is doing right now in a conversation, for example after a request with the wrong details.

Input

Field	Type	Description
conversation	string	The conversation's friendly name. Can be omitted when there is only one.

Output

A confirmation that the cancel was requested, or a note that nothing was running.

Example

You say: “Stop, I gave Bridget the wrong market.”

json

{ "conversation": "market sizing chat" }

Documents#

bridget_request_documents#

What it’s for

Getting you a secure link to upload your files to Teamroom. The files can support one request, become lasting sources in the project, or go into a project being created.

Input

Field	Type	Description
intent	message \| project \| create_teamroom	What the files are for: one request (the default), the project’s lasting sources, or a new project.
teamroom	string	Project the files belong to.
purpose	string	Short description shown on the upload page.

Output

The upload link and an upload ticket id. The link is valid for 7 days. If you upload straight away, the result lists the uploaded files instead.

Example

You say: “I have three market reports to add to <project name>.”

json

{
  "intent": "project",
  "teamroom": "<project name>",
  "purpose": "Market reports for the project"
}

bridget_await_documents#

What it’s for

Waiting while you upload through the link. The assistant repeats it until your upload is finished.

Input

Field	Type	Description
ticketIdRequired	string	The upload ticket id.

Output

The list of uploaded files once you are done. Until then, it says it is still waiting and shows any files uploaded so far.

Example

You’re still uploading when the first wait ends:

json

{ "ticketId": "<upload ticket id>" }

bridget_add_project_documents#

What it’s for

Adding uploaded files to the project as searchable, citable sources for all future conversations, interviews and deep research.

Input

Field	Type	Description
uploadTicketIdRequired	string	The upload ticket id, once the upload is finished.
teamroom	string	Project name. Not needed if the upload link was made for a specific project.

Output

A confirmation of how many files were added and to which project, with the file names.

Example

After you finish uploading the market reports:

json

{ "teamroom": "<project name>", "uploadTicketId": "<upload ticket id>" }

Outputs#

bridget_read_artifact#

What it’s for

Opening a note or deliverable Bridget produced, such as an interview guide, a research report or a synthesis.

Input

Field	Type	Description
artifactRequired	string	The artifact's name, or the link to it from Bridget's reply.

Output

The full content as text. For files without text content, such as a slide deck, it returns a fresh download link.

Example

You ask: “Show me the full interview guide Bridget wrote.”

json

{ "artifact": "<artifact name>" }

Reference#

Look up as needed.

Access and security#

You act as yourself. Each connection signs in as the person who authorized it. The assistant can do only what that person can do in Teamroom.
Your projects only. Bridget sees only projects you are a member of, and each project's compliance template still applies.
Removing access. To disconnect, remove the connector or server in your client. Removing someone from a project in Teamroom also removes their access to it through MCP.
Files go straight to Teamroom. Uploads go through the secure upload link, not through the assistant.

Timing and limits#

Item	Behavior
Wait inside `bridget_send_message`	About 60 seconds, then the tool reports that Bridget is still working.
Each `bridget_await_response` call	Waits up to about 45 seconds. Call it again to keep waiting.
Long tasks	Deep research usually takes about 50 minutes (80 at the 90th percentile). Recruited interviews take days. Work continues in Teamroom after the chat ends.
Upload links	Valid for 7 days. Each upload is used once. Ask for a new link for more files.
Download links	Short-lived. Download files when you get them, or open them in Teamroom.
Conversation list	Up to 50 per call; 20 by default.
Pricing	MCP has no charge of its own. Bridget’s work uses Teamroom credits, the same as in the web app.

Troubleshooting#

Bridget’s tools don’t appear.

Why: The server wasn’t added as a remote HTTP server, the URL is wrong, or the tools are turned off in the chat.

What to do: Check the URL exactly, then reconnect. Make sure Bridget is turned on in the chat’s tools menu.

Sign-in fails, or the client asks you to sign in again.

Why: The sign-in didn’t finish, or the session expired.

What to do: Reconnect and sign in again with your Teamroom account. If you have several accounts, check you chose the right one.

The assistant answers without using Bridget.

Why: The prompt didn’t make it clear that Bridget should do the work.

What to do: Say “Bridget” or “Teamroom” in the prompt and name the project.

A project is missing or “not found”.

Why: You aren’t a member of it, or the name doesn’t match.

What to do: Ask the assistant which Teamroom projects you can use. Ask a project admin to add you if needed.

“Bridget is still working.”

Why: Normal for long tasks.

What to do: Ask the assistant to keep waiting, or come back later and ask it to check the conversation. Don’t resend the request.

The client times out on a long task.

Why: Some clients limit how long one tool call can run.

What to do: The work continues in Teamroom. Ask the assistant to check the conversation, or open it in Teamroom.

No form appears for Bridget’s question.

Why: Your client doesn’t support MCP forms (Claude web and desktop, for example).

What to do: Answer in the chat. The assistant passes your answer to Bridget.

“That upload link has expired.”

Why: Upload links are valid for 7 days.

What to do: Ask for a new upload link.

Bridget can’t cite a document you shared.

Why: The file was pasted into the chat instead of uploaded.

What to do: Upload it through the secure link (see Documents).

A tool is missing after an update.

Why: Your client is still showing the old list of tools.

What to do: Reconnect the server, or restart the client.

For anything else, email support@teamroom.ai. Include your client and its version, the project name, and the time of the problem.

Was this page helpful?