Maven Bio Plugin
The Maven Bio plugin is a primitive-first research and evidence layer for BioPharma
intelligence workflows. It supports investment teams, pharma BD and corp dev, and in-house
portfolio strategy across drug pipelines, clinical trials, regulatory events,
public-company financials, and private financing rounds.
Requirements
An active Maven Bio account. The plugin connects to the Maven Bio MCP connector at
https://mcp.mavenbio.com, which is read-only and scoped to your own account and
organization. Authentication uses OAuth; no API keys or passwords are shared with your
client.
- Documentation: https://mavenbio.com/mcp/docs
- Privacy policy: https://mavenbio.com/mcp/privacy
- Support: support@mavenbio.com
What it provides
- research primitives under
skills/covering competitive pipelines, indication
landscapes, asset and company profiles, market sizing, deal and financing activity,
commercial analogs, and evidence synthesis - approval, label, and exclusivity analysis anchored in primary FDA and SEC filings
through document search and read - a specialized research sub-agent under
agents/ - MCP connectivity via
.mcp.json
Hermes Agent
This directory is also an Agent Plugins v1 package (plugin.json, mcp.json, and
skills/) for Hermes Agent. It holds configuration and Markdown only, with no Python or
executables.
hermes mcp install maven-bio # signs in to Maven Bio through OAuth in your browser
hermes plugins install maven-bio # adds the research workflow skills
hermes plugins enable maven-bio
What to know before installing:
- Every tool call goes to the hosted Maven Bio MCP server at
https://mcp.mavenbio.com. - All tools are read-only and scoped to the signed-in account and organization.
- Sign-in uses OAuth with a local callback. No API keys or passwords are stored in
config, and Hermes stores and refreshes the token. - The plugin sends no telemetry of its own.
Design principles
- structured MCP tools define the baseline universe
- document search and reads augment and reconcile master analyses
- evidence is attached as structured data, not custom citation markup
- canonical source references are used instead of
srcNaliases - gaps are surfaced explicitly rather than smoothed over
- Claude built-in sub-agents are preferred over async task tools for interactive
workflows - the plugin does not ship hooks, custom runtime state, or artifact-format opinions; those
belong to the host environment or the calling workflow
Instruction model
Runtime behavior comes from the supported plugin surfaces:
skills/*/SKILL.mdagents/*.md- MCP tool descriptions served by the connector
.claude-plugin/plugin.json
Changelog
0.6.0
- Approval, label, and exclusivity questions now route through document search and read
against primary FDA and SEC filings rather than through dedicated regulatory tools. - Analysis scaling guidance: when a set is larger than the evidence budget, narrow the set
rather than thinning the evidence gathered per row.
README from the reviewed commit e77a5d0; it updates when the author re-pins.

Maven Bio