Field-by-field comparison

Compare the evidence before choosing an MCP.

Each signal stays separate, missing facts remain visible, and the column order follows your selection.

MCPs in this comparison

Citra (PDF Reader MCP)Codebase Memory

Uncheck an MCP to remove it. Comparing two or three keeps each signal in its own row.

Comparable alternatives

Azure MCP ServerMCP100 SelectedChrome DevTools MCPMCP100 SelectedContext7 MCPMCP100 SelectedFirecrawl MCPMCP100 SelectedGrafana Cloud MCPMCP100 SelectedGoogle MCP Toolbox for DatabasesMCP100 Selected

Shown only when capability or industry context overlaps.

Side-by-side MCP evidence comparison
Evidence field	Citra (PDF Reader MCP)Watchlist candidate	Codebase MemoryMCP100 Selected
Task fit
Best for	Reading and searching PDFs with structured text and tables OCR or rendered PDF evidence with page and bounding-box citations	Repository indexing, symbol search, call-path tracing, impact analysis, and architecture queries across many languages
Not ideal for	Editing, creating, signing, or rewriting PDF files, which are outside the documented three-tool surface Word, Excel, CSV, or general document workflows outside PDF	Tiny repositories, one-off text searches, or semantic questions unsupported by the indexed parsers
Avoid when	The process can read sensitive PDF directories that have not been explicitly allowlisted You cannot review the Node and platform-native package chain before installation	The source tree cannot be indexed locally or persistent cache files are prohibited
Provenance
Provenance details	First-party MCP SylphxAI · Publisher source ↗	Community implementation io.github.DeusData · Publisher source ↗
Maintenance
Maintenance details	Repo: Aug 8, 2026 Package: Aug 7, 2026	Repo: Aug 14, 2026 Package: Aug 14, 2026
Popularity evidence
GitHub stars	891 GitHub stars · checked 2026-08-14T18:46:26.000Z	38,935 GitHub stars · checked 2026-08-14T18:46:26.000Z
30-day package downloads	No reliable download data	28,015 30-day package downloads · checked 2026-08-14T18:46:26.000Z
External adoption evidence	github: repository stars: 891Exact attributable signal.github: repository stars · Admission snapshot · checked 2026-08-14T18:46:26.000ZEvidence source ↗	github: repository stars: 38,935Exact attributable signal.github: repository stars · Admission snapshot · checked 2026-08-14T18:46:26.000ZEvidence source ↗
Client coverage
Client coverage details	Claude Code✓ Local stdio Claude Desktop✓ Local stdio Codex✓ Local stdio VS Code✓ Local stdio Cursor✓ Local stdio OpenCode✓ Local stdio	Claude Code✓ Local stdio Claude Desktop✓ Local stdio Codex✓ Local stdio VS Code✓ Local stdio Cursor✓ Local stdio OpenCode✓ Local stdio
Client coverage note	Local stdio runs on your computer. Streamable HTTP connects to a remotely hosted MCP server.
Access and data
Authentication	Default stdio needs no service credential. Optional HTTP mode supports MCP_API_KEY and binds loopback by default; configured HTTP or OpenAI-compatible visual providers may use their own headers or API key, while local command providers do not inherently require one.	No authentication or API key is documented for local stdio. CBM_ALLOWED_ROOT is an optional filesystem containment control; no containment is imposed when it is unset.
Cost	The MIT-licensed local server has no documented usage fee. Pricing for any operator-selected external OCR, visual, network, or client provider is outside the Citra documentation and remains provider-specific.	The MIT-licensed native software lists no service fee. Local storage and package-manager costs are separate.
Permissions	The tools read PDF paths and URLs available to the process. Directory allowlists can restrict local reach; URL handling blocks private addresses by default unless the operator overrides it.	Indexes repositories, queries code and graphs, performs architecture and impact analysis, manages ADRs and watchers, deletes projects, and writes agent configuration, instruction, Skill, and hook files.
Data handling	Default stdio processing is local. Optional OCR or visual providers may receive rendered content according to their configuration, and URL inputs are fetched from their source.	The publisher says processing is local and code does not leave the machine. Indexes, embeddings, memories, SQLite data, and logs persist under the local CBM cache directory.
Limitations
Tradeoffs	Local-first PDF processing reduces routine cloud exposure, but optional OCR and visual providers introduce separate dependencies and data paths. The focused three-tool surface is easier to reason about than a general file server, but it deliberately does not cover document editing.	Persistent indexing can reduce repeated reads, but consumes local storage and graph results can become stale until the repository is re-indexed.
Risk context	Use stdio, configure explicit allowed PDF directories, retain private-IP URL blocking, require an API key before any non-loopback HTTP deployment, and review native artifacts before installation.	Indexing can capture sensitive files, while persistent graph data and installer/configuration writes expand local exposure. Set CBM_ALLOWED_ROOT, protect the cache, and use explicit project paths.
Evidence date
Editorial review	2026-08-14	2026-08-11
Candidate evidence	2026-08-14T18:46:26.000Z	2026-08-14T18:46:26.000Z

Popularity, maintenance, fit, permissions, and client support are independent evidence fields.

Read the evidence method →