Research for Web3 builders and AI agents
Give your agent research it can read.
Find relevant papers and protocol specifications. Read the methods, equations and assumptions. Connect your next engineering decision to source evidence.
Start here · judging walkthrough
One question. A source you can inspect.
Connect dtox to your MCP-compatible agent, then ask it to investigate an engineering question. Research reads are free, with no signup, API key or wallet required. The tools run inside your connected agent.
Claude Code
claude mcp add --transport http dtox https://read.whoim.space/mcp
Codex
codex mcp add dtox --url https://read.whoim.space/mcp
Other HTTP MCP clients · server URL
https://read.whoim.space/mcp
Ask your agent
Find research on data availability sampling for blockchain light clients with limited bandwidth. Open a relevant paper, read its method and assumptions, and explain the implementation trade-offs with source citations. Distinguish the authors' findings from your own conclusions.
- Discover. Use
find_papersorsearch_research_paperto find candidate sources. - Inspect. Open a result with
get_paper, then read the relevant method withread_paper_section. - Check the detail. Retrieve equations or algorithms with
get_code_or_math_spec, or collect a cited packet withget_research_bundle.
A quick starting point: ask for paper 2005.11401, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, then read its Introduction. Its card, outline and section read were checked through the public MCP on 17 October 2026.
Look for the source link, the actual passage and a clear connection between the evidence and the agent's conclusion. See the research workflow.
What you can do
From a promising abstract to an engineering decision.
Useful knowledge is scattered across papers, specifications and technical documents. An implementation often depends on an equation, a security assumption or a limitation buried in a later section. dtox brings those sources into the agent's workflow.
Read the method
Open source sections, equations, algorithms, code and tables where available. Inspect the evidence behind the answer.
Compare approaches
Explore related work, follow citation links and compare two methods' reported results and assumptions.
Investigate an idea
Pass distinct technical claims to validate_project. Review the candidate papers and read the passages supporting each claim.
Another way to try it
Use validate_project to investigate a retrieval-augmented language model. Evaluate this claim separately: "Retrieval augmented generation conditions language-model answers on retrieved documents." Show the candidate papers and supporting passages, then explain what still needs verification.
Use dtox to investigate data availability, account abstraction, protocol incentives and agent research systems. Responses identify the source and available coverage, including full text and abstract-only records.
Why Solana
Agent research with attributable readings.
Solana devnet integrationWallet-signed readings · SAS
An agent signs a canonical message containing the claim, paper and verdict. The attestor verifies the reviewer's signature and records the reading through Solana Attestation Service. The signature identifies who signed; the source evidence supports the reading.
Explore the attestorAgent payments · x402
A separate MCP gateway implements per-call USDC payments for research tools and signed readings on Solana devnet. Builders can explore this integration alongside the free research endpoint.
Explore the payment gatewayRead source evidence → prepare a verdict message → agent wallet signs → verify signature → SAS attestation
The research walkthrough above uses the free endpoint. A wallet is used when signing a reading; payment and signing details are available in the integration guides.
How it works
Find the paper. Retrieve the passage. Follow the source.
Semantic retrieval
Paper-level search shortlists relevant documents, then passage-level search finds the useful detail within their contents.
Exact technical terms
Lexical search adds precise terms, identifiers and protocol names to the semantic retrieval channels.
Citation links
Citation neighbours provide another route to related work. Results carry source links and metadata for inspection.
Built with FastAPI, Qdrant, SQLite FTS5 and ONNX embeddings, exposed through MCP. Read the architecture overview.
13 public MCP tools
A research workflow inside your agent.
find_papers
Discover papers and specifications.
search_research_paper
Find indexed passages for a question.
get_paper
Inspect a document card, outline and citations.
read_paper_section
Read a selected source section.
count_papers
Explore corpus counts and topic facets.
similar_papers
Find semantically related documents.
get_code_or_math_spec
Retrieve equations, algorithms, code and tables.
compare_methods
Inspect evidence from two documents.
research_trends
Explore topics and publication trends.
validate_project
Investigate distinct technical claims.
get_research_bundle
Collect a bounded packet of cited evidence.
get_verdict_message
Prepare a canonical message for signing.
record_signed_verdict
Submit the caller's wallet-signed reading.
The research index
Web3 research and the technology around it.
Four layers cover Web3, AI agents, language models and builder technology. Sources include arXiv, IACR ePrint, ACL Anthology, PMLR, open-access research, Ethereum EIPs/ERCs, Solana SIMDs and technical whitepapers.
Corpus counts checked 17 October 2026, 15:48 UTC.