MCP server
Ask about a study from Claude, and get the real answer.
Connect Claude Desktop, or another MCP client, and ask about an engagement or a benchmarking study in plain words. Each answer comes from the same services the app uses, under your own sign-in and role.
In one line
A question in, a figure from your study out.
The client picks the tool. TAIGA answers from the study, and the client puts it in words.
What it can do
67 tools, grouped by the work.
Read and act on engagements, entities, documents, the IRD wizard and benchmarking studies. 10 resources let a client read an engagement or a study directly, and 8 guided prompts run common jobs like reviewing an extraction or comparing studies.
- Local File generation and compliance checks are not available here yet.
Active tools by area
67 in all
- Standalone benchmarking studies17get_study_alp
- Document extraction8trigger_extraction
- Engagement benchmarking8get_benchmarking_status
- IRD wizard7get_wizard_progress
- Documents7upload_document
- Entities4list_entities
- Engagements4list_engagements
- Submission and approval4check_submit_readiness
- Guidance and dashboard4suggest_next_steps
- Teams3list_my_team_members
- TP calendar1list_upcoming_deadlines
No second copy
Every call goes through the app’s own services.
The MCP server holds no data and keeps no session. Each request is checked, then turned into a call to the same endpoints the web app calls, carrying your firm, user, role and perspective.
Your MCP client
POST /mcp with your sign-in token
TAIGA MCP server
Verifies the token, checks the tool’s roles, keeps no session and no data
The app’s own services
GET /api/v1/benchmarking/studies/{studyId}/alp
Your firm’s data
Scoped to your firm, the same as in the app
Carried on every call
- Firm
- User
- Role
- Perspective
Roles
Each tool names the roles that may call it.
A tool with no role list is refused by default. Firm admins can call every tool. A refusal comes back as a plain message saying which roles the tool needs.
Sign-in
Claude Desktop finds the sign-in on its own.
The server publishes OAuth 2.1 protected-resource metadata, so a client learns where to sign in without setup. Every call then carries a token that is verified before any tool runs.
GET /.well-known/oauth-protected-resource
{
"resource": "https://<your TAIGA API>/mcp",
"authorization_servers": ["https://<your sign-in host>"],
"scopes_supported": ["mcp:tools", "mcp:resources", "mcp:prompts"],
"bearer_methods_supported": ["header"]
}- 1The client reads where to sign in.
- 2You sign in there, and the client receives a token.
- 3Each call carries your token, checked before any tool runs.
- 4No token, or a bad one, gets “Authorization required”.
Coverage
What it covers.
- Tools
- 67 active
- Grouped across 11 areas
- Resources
- 10
- Engagements, entities, extractions, benchmarking
- Prompts
- 8
- Guided multi-step jobs
- Transport
- Streamable HTTP
- No session kept between requests
- Sign-in
- OAuth 2.1 discovery
- Protected-resource metadata, RFC 9728
- Access
- Denied by default
- Each tool lists the roles that may call it
Questions
Before you ask.
- Does it keep its own copy of our data?
- No. It holds nothing between requests. Every answer is read from the same services and firm-scoped data the app uses.
- Can a taxpayer run benchmarking through it?
- No. A taxpayer can list their engagements, upload documents and submit their IRD. Running a search or overriding a comparable needs a practitioner role.
- Which clients can connect?
- Claude Desktop, through OAuth discovery, and any MCP client that speaks MCP over HTTP with a bearer token.
- Can it generate the Local File?
- Not yet. That tool is listed, but it answers that document generation is not available. Generate reports in the app.
