Deedfox
Follow the evidence across documents and turn your findings into reports you can review and use.
A document points to another document. Names change, dates conflict, and missing records leave questions unanswered. Deedfox helps your agent follow references, organize evidence, and turn findings into the document you need.
Connect it to your AI agent, choose a case, and explain the question you want answered and the document you want back. The service runs the examination on the operator’s server; your agent can retrieve its reports and use them to draft a different deliverable. Choose your output for examples.
Connect once, then ask
You need the operator’s MCP URL and private access key, plus Node.js 22.16 or newer and npm. Have Claude Code, Codex, or another stdio-capable MCP client installed. You do not need Python, local survey PDFs, or model-provider API keys.
macOS or Linux
curl -fL https://deedfox.ayushworks.com/downloads/setup_mcp.mjs -o deedfox-setup.mjs
node deedfox-setup.mjs
Windows PowerShell
Invoke-WebRequest https://deedfox.ayushworks.com/downloads/setup_mcp.mjs -OutFile deedfox-setup.mjs
node deedfox-setup.mjs
The wizard asks for your URL and hides your key while you type. It installs the pinned bridge in ~/.deedfox, stores the key there, and offers to register Claude Code or Codex. For another client, it prints the command to add in that client’s MCP settings. Existing kits and server entries are preserved. See connection options and setup recovery if you prefer direct HTTP or already have a connection.
Restart your agent session and approve the MCP connection if prompted. Then ask:
Use Deedfox to discover its tools and list the available surveys.
Do not start a paid examination yet.
Success: your agent can discover six tools and call list_surveys. If it cannot, follow Troubleshooting. Listing surveys does not invoke title models.
Your first case
Choose an exact survey returned by the service. A full examination uses the operator’s paid model credits. Agree on the case and spend before sending this request:
Read https://deedfox.ayushworks.com/llms.txt.
Examine <exact survey name> using the prepared records, with browser sourcing
and purchases disabled. I authorize this case's model spend.
Save the job ID and follow its status. Retrieve the available reports and
show me the source-backed findings, missing evidence, and review priorities.
Your agent starts one job, saves its ID, and retrieves the reports when ready. The job continues if the client disconnects; reconnect using that ID rather than starting again. Your first examination explains the full workflow.
flowchart LR
accTitle: From prepared case to review
accDescr: Choose a registered survey, start one job and keep its ID, retrieve available reports, then review the original records.
A["Choose survey"] --> B["Start once<br/>Keep job ID"]
B --> C["Read reports"]
C --> D["Review records"]
What comes back
Ask your agent to turn the findings into the format you need: a chronology, a research brief, a comparison, or a checklist. The table below describes the built-in title workflow, which is the worked example used throughout these guides.
| Deliverable | What it helps you do |
|---|---|
| Lease Run Sheet | Review the instruments, parties, dates, acreage, and recorded references in order. |
| Proposed title opinion | Trace ownership and inspect the title notes and curative requirements. |
| Ownership and buying summary | Check fractions, net mineral acres, royalty burdens, and unresolved conditions. |
| Review and faithfulness reports | Find corrections, unsupported extractions, and evidence gaps. |
See a fictional report and worked ownership example. The illustrations are fictional, decorative scenes, not outputs or evidence from an examination.
Where it helps, and where it stops
Define the question and required evidence for each case. The guides use title research as a worked example. The operator supplies the case records and context; the MCP service has no upload tool. Arrange access to the original scans separately for review.
You can ask your connected agent for a different document using the returned findings. Changing the underlying extraction fields, legal rules, or built-in report structure is an operator/developer task. See adapt the workflow for where prompts, schemas, and providers fit together.
It is an examination assistant. A finished job and a balanced fraction total do not establish legal title or approve a mineral purchase. Missing instruments and operative clauses still need an examiner. Read the measured accuracy and limits before relying on a result.
Keep going
- Connect any compatible agent, through authenticated HTTP or the local bridge.
- Try the arithmetic without running OCR.
- Review the results and give useful pilot feedback.
- Read the six-tool contract or develop the engine.
Agents can start with /llms.txt or read /llms-full.txt. The documents explain the tools; access still requires the operator’s private key.