Research-to-report workflow for agents: plan sources, score them for freshness and conflicts, and reject claims without citations.
Agent-written reports often sound sure of themselves and cannot say where a single sentence came from. Research Scope attacks that directly. Instead of trusting the model to cite well, the Research Scope MCP server makes scoring and citation coverage mechanical: sources get graded, reports get a numbered registry, and a final check fails any claim that points at nothing. The goal is a report a human can audit in about a minute and a half.
plan_research takes a scope, plus optional domain hints, and returns a source-target plan with kill criteria. A typical one: fewer than one primary source and two independent secondaries means stop and report insufficient evidence.fetch_and_score fetches one URL and returns a 0-100 score with flags for primary status, freshness, paywalls and conflicts of interest, each with its evidence. Scored sources are registered for later use.build_report builds a report shell under data/reports/ with a claims section, [Sn] citation numbering and a scoring appendix. Only already-scored sources are accepted.check_report returns pass or fail plus a list of orphan claims.| Dimension | Max |
|---|---|
| Availability and substance | 20 |
| Primary-source status | 32 |
| Freshness | 24 |
| Access (paywalls, registration) | 12 |
| Independence (sponsored, affiliate) | 12 |
Freshness earns full marks within 30 days and decays to 2 points past five years, or zero if no date can be found. Primary status leans on host class, such as government, education or filings, and on press or announcement paths. Grades follow the usual cut lines: A from 80, B from 65, C from 45.
Run it over stdio with npx research-scope-mcp, or build locally with npm ci && npm run build. The global npm install applies once the package is published, so a local build is the dependable route until then. A Streamable HTTP mode (--http --port 3001) exists for remote deployments; unauthenticated HTTP requests get a 401 with discovery headers. A Dockerfile is included as well.
Scoring in the Research Scope MCP server is heuristic. Primary status is inferred from the host class, so a government summary page may still score as primary, and a site without a detectable date loses all freshness points. fetch_and_score needs network access, and it scores the page you hand it, not the claim you intend to support. A paid kit with guides and examples is offered separately; the server is MIT.
Compare it with other integration servers, or browse the rest of the Other Integrations listings.
Check the linked publisher documentation and your connected client's tool list. Names and availability can vary with the server version and account permissions.
| Tool | What it does |
|---|---|
| plan_research | Turn a research scope (and optional domain_hints) into a source-target plan with kill criteria. |
| fetch_and_score | Fetch one source URL and return a 0-100 score with primary, freshness, paywall and COI flags. |
| build_report | Assemble a citation-registry report shell from already-scored sources. |
| check_report | Check that every substantive claim cites a scored source; returns orphan claims. |
{
"mcpServers": {
"research-scope": {
"command": "npx",
"args": [
"research-scope-mcp"
]
}
}
}Runs over stdio. If the npm package is not yet published, build locally with npm ci && npm run build and run the built server instead.
Run an Actual Budget ledger from your AI agent, with phone approvals for every change on paired machines.
An Odoo module that serves a read-only MCP endpoint at /mcp, scoped to each user's own access rights.
Drives DaVinci Resolve 21.1, free edition included, from Claude Code through a Lua bridge running inside the app.
Agent memory that keeps every correction with its history, plus saved working state so a crashed run can resume.
Scans a Windows screen into a numbered element table so an agent clicks and types by id instead of guessing pixels.
Manage a Linux VPS, Postgres and Odoo databases and GitLab deploys over SSH, with previews, allowlists and an audit trail.