MCP server.
Seven tools for running Tagsight extractions and pulling structured results from any MCP-compatible desktop client. Python pip install, API-key auth, same quota as the REST API.
The seven tools.
Each tool is a single MCP function the client invokes by name. Only extract_io_list consumes a credit; polling and read tools do not.
| Tool | Purpose | Credits |
|---|---|---|
extract_io_list | Start an extraction on a PDF or image. Returns a job_id. | 1 / call |
get_job_status | Poll extraction progress by job_id. | No |
get_job_results | Fetch the full structured output once extraction completes. | No |
list_jobs | List recent extractions for the authenticated account. | No |
get_io_summary | Signal-class counts (AI / AO / DI / DO) and BPCS / SIS split. | No |
search_instruments | Find tags across the account by pattern, signal class, or service. | No |
ask_tagsight | Ask a natural-language question against your extracted instrument data. | No |
Client configuration.
Add this block to the MCP server config of the desktop client. The exact filename and location vary by client; consult the client's MCP settings docs for the path.
{
"mcpServers": {
"tagsight": {
"command": "tagsight-mcp",
"env": { "TAGSIGHT_API_KEY": "ts_your_key" }
}
}
}Example prompts.
Phrasings that work with most MCP-aware desktop clients. The client picks the appropriate tool from the catalogue and passes the parsed arguments to the server.
Extract the I/O list from ~/drawings/FS-101.pdf and give me the signal-class totals.Poll job_id abc123 until it completes, then summarise the instrument count by signal class.List my last ten extractions and pull the structured results for the most recent one.Run extractions for every PDF in ~/drawings/area-300 and keep the job_ids.Summarise the instrument count by signal class for the most recent extraction.
Authentication, security.
- API key in environment variable only, never in config files checked into source control.
- Per-machine keys, scoped, revocable individually from the dashboard.
- Same rate limits, same authentication, same data controls as the REST API.
- Results are not stored locally. MCP runs against the same account data as the web app.
Common questions.
- What is MCP?
- Model Context Protocol, an open standard for exposing tools and data sources to desktop AI clients. An MCP server advertises tools; an MCP client invokes them. Tagsight publishes an MCP server so any MCP-compatible client can run extractions and pull structured results.
- Which clients can use the Tagsight MCP server?
- Any MCP-compatible desktop client. A current client will have MCP support documented in its settings or extensions panel.
- Do I need the Tagsight web app to use MCP?
- A Tagsight account and an API key. Extractions behave exactly like web-app extractions: same account, same credits, same outputs. The MCP client is just a different surface.
- Is MCP included in my plan?
- MCP access is included on paid plans. Extractions run through the MCP server count against your drawing quota exactly like web-app extractions. The polling and summary tools do not consume credits.
- How does authentication work?
- The API key is passed via environment variable. The MCP server authenticates every request. Same key, same scopes, same rate limits as the REST API.
Run an extraction from the desktop.
An API key, the pip install, and a single MCP block in the client config. Same quota as the web app and the REST API.