Documentation
MCP Connector
What you can do
The connector exposes your advisory work as a small set of tools your assistant can call. It is read-mostly: most tools are quick lookups across projects, scope and deliverables. The assistant can also act on your behalf: write_claims authors claim highlights on a PDF or Word document, add_feedback_comment leaves feedback comments on a PowerPoint, and request_feedback kicks off an AI feedback run. It can also work your project tasks: planning a task out as checklist steps, reporting progress as it goes, and creating sub-tasks under a task you set up. Everything runs under your account, so the assistant only ever sees projects you are a member of.
Requirements
- A LithicBlue account on a paid plan (the connector requires a paid-tier entitlement).
- An MCP host that supports custom connectors, for example Claude or ChatGPT developer mode.
Connect your assistant
Add a custom connector in your AI host pointing at the LithicBlue MCP endpoint. Use your regional URL (the same host you sign in at) and append /api/mcp/mcp to it:
https://eu.lithicblue.com/api/mcp/mcpUse the host you sign in at
LithicBlue runs on regional deployments, for example eu.lithicblue.com, na.lithicblue.com or au.lithicblue.com. Point the connector at whichever one your firm uses; swap the example host above for yours.
- In your AI host, add a new custom connector / MCP server.
- Paste the endpoint above. The host auto-discovers the connector and registers itself automatically.
- You will be sent to LithicBlue to sign in (OAuth).
- LithicBlue then shows an Approve access screen naming the assistant, the host it connects back to, and what it will be able to do: read your projects, deliverables and claims, and act on your behalf by authoring claims, adding feedback comments, starting feedback runs and updating your project tasks. Approve to connect it to your account, or decline to stop there; nothing is shared unless you approve.
- The LithicBlue tools appear in your assistant, ready to use.
Authentication is per-user
The connector signs you in with OAuth and acts as you; it never grants access to a project you are not a member of, whatever an assistant asks for.
Assistant says it has no access to LithicBlue tools?
Tool access is its own permission on the Approve access screen. If your assistant reports that the connection was not granted access to LithicBlue tools, disconnect LithicBlue in your assistant, connect it again, and approve.
Only assistants we support can connect
LithicBlue connects only to the assistants listed on this page. If the Approve access screen names something you do not recognise, or you did not just start a connection yourself, decline it, and let us know.
Prefer step-by-step instructions for your assistant? See the dedicated guides for ChatGPT, Claude, Microsoft Copilot and Gemini.
Available tools
Start by asking your assistant to list your projects; the other tools work against a project from that list.
| Tool | What it does |
|---|---|
list_projects | List the projects you can access. Call this first to get a project to work with. |
get_scope_status | Read a project's scope analysis: per-section fulfilment, or one section's requirements, answers and sources. |
get_data_gaps | Read a project's data-gaps analysis: what it still needs, overall or for one section. |
list_deliverables | List a project’s deliverables (PowerPoints, PDFs and Word documents) with their file type, size and whether verification has run. |
read_deliverable | Read a deliverable’s text, page by page or slide by slide, in order. |
get_claims | Read stored claim-verification results: a verdict breakdown across the whole document first, then any single slide, page or claim in full with the sources behind it. |
write_claims | Author claim highlights on a PDF or Word deliverable, anchoring each claim to a verbatim span with sources. |
list_feedback_personas | List the feedback personas available for a deck: the built-in roles and your own custom personas. |
request_feedback | Run AI feedback on a PowerPoint with your chosen personas. Runs in the background; the comments appear in the deck’s feedback panel. |
read_feedback_comments | Read the AI feedback comments on a PowerPoint: each with its persona, severity, finding and suggestion. |
add_feedback_comment | Leave a feedback comment on a slide of a PowerPoint, with a severity and an optional suggestion. |
list_my_tasks | List your tasks in a project, optionally by status, so the assistant can find the one you asked it to work on. |
get_task | Read one task in full: its description, status, checklist steps, recent activity and sub-tasks. |
add_task_steps | Lay out a plan as checklist steps on a task, so you can see what the assistant intends to do before it starts. |
update_task_step | Tick off, skip or reopen a checklist step as the assistant works through it. |
update_task_progress | Change a task’s status or leave a progress note, attributed to the assistant, as work happens. |
create_task | Break a task you created into sub-tasks. Assistants can only add sub-tasks under an existing task, never new top-level work. |
Example prompts
- “List my LithicBlue projects, then summarise the open data gaps on the Acme diligence.”
- “Read deliverable 3 in that project and tell me which claims are still unverified.”
- “Show me only the contradicted claims on the Acme deck, then walk me through the evidence behind the worst one.”
- “On the Acme PDF, anchor a verified claim to the revenue figure on page 4 and cite the source I gave you.”
- “Run writing and structure feedback on the Acme current-state deck.”
- “Pick up the ‘Market sizing’ task on the Acme project, plan it out as steps and update the task as you go.”
A note on writing claims
When an assistant calls write_claims, each highlight is anchored to a verbatim span of a page and carries a verdict (verified, contradicted or unverified) plus the sources behind it. These are advisor-authored and pinned, and show up in the deliverable just like claims you write yourself. Review them as you would any AI output before sharing with a client.
A note on AI feedback
request_feedback starts the feedback run and returns straight away; the personas work in the background for a minute or two, and their comments land in the deck’s feedback panel in LithicBlue rather than coming back through the assistant. Open the deck (or ask the assistant to read_feedback_comments later) to see the findings.
A note on tasks
An assistant can plan, progress and break down tasks, but it cannot start new top-level work: create_task only adds sub-tasks under a task a person created, so the plan always begins with you. Steps it ticks off and progress notes it leaves appear on the project’s task board as they happen, attributed to the assistant, so you and your team can follow along without asking for a status update. Review its work as you would any AI output before treating a task as done.